Home

Reducing False Alarms with Multi

1. i i y 60 cy 300 a A h lt l 40 3 200 8 A20 100 i i 0 L o ti t2 t3 ta ts te tr 3 00PM _ 3 30PM 4 00PM 2 time amp on 3000 C e 5 E i e gt 8 off G A 0 i time time Figure 14 A stable with fluctuation example for robust ness analysis e signal k 4 0 1 80 D S fon w Q pseudo ADC D mote ADC N S 3S Ny 100 o ty t2 t3 ta ts te ty o on 3000 v pseudo P Cc off time time Figure 15 A monotonically increasing example for ro bustness analysis erence value and hence Cy cannot easily build up Our al gorithm is capable of suppressing false alarms in this case showing in Figure The final case is when the pipe skin temperature mono tonically increase The physical meaning is that fluid flow resumes after temporary pump shut in We expected that our algorithm fall back to silent quickly after temperature be comes high enough Similar to the stable case we denote the temperature readings over time as Si Si 2 and s gt Si 1 i 2m 1 Likewise the reference value k between t ti 1 Si 1 Sj 0 1 _ Hi _ 5 lt Si lt Si41 Since k is always smaller than the temperature Cy will not build and hence our algorithm achieves true negative This process reversely mirrors what we observe in the
2. pseudo P time time Figure 13 A monotonically decreasing example illus trates the robustness of our blockage detection algorithm half of Figure 13 illustrates this process The right half is an excerpt of our experiment data rendering similar behav ior The k solid thick line in the top right plot comes from the result of our automated algorithm On the contrary the k dashed line comes from the hand set version to show what we expected to see if parameter setting is perfect which yields C below The second case is when the pipe skin temperature gen erally stabilized but with natural fluctuation The physical meaning is that fluid flow normally We expected that our algorithm should remain silent because temperature should stay high enough Although this model can represent the sit uation when pipe completely cools off we do not discuss it because of no equivalent data We denote the temperature readings over time as Si Si 2 and without losing generality Si gt Si 1 i 2m 1 The quality and anomaly level update schema is the same as the last case which leads to the reference value k between ti ti 1 l i 2m 1 i 2m Si 1 Sj 2 lt Sj gt Si The above relationship between reference value and signal indicates that the temperature is likely to envelope the ref 2 ki e signal k 4 0 1 80 D S
3. We next introduce the three different temperature models monotonically decreasing stable with fluctuation and monotonically increasing The first case we look into is when the pipe skin temper ature monotonically decrease The physical meaning behind this model is that fluid flow stopped either because of pump off or blockage Under this situation we expected that our flow presence algorithm based on one sided CUSUM should trigger because observed temperature signal should stay be low the reference value significantly long enough We denote the temperature readings over time as i 2m 1 i 2m Si gt Si 1 According to the pumpjack status above and without losing generality we assume pumpjack switches from on to off at tom At the same time we update the quality level Mn cor respondingly i 2m 1 i 2m Please see Section 2 3 for more details about parameter tun ing Likewise at t241 we update the anomaly level u2m 1 i 2m 1 i 2m Hence the reference value k between t t 1 is wt _ sits 2 2 This result clearly show that the reference is always larger than immediate temperature observation and therefore cer tainty of drop C4 builds up triggering algorithm The left ki gt Si gt Si41 signal k 0 v 1 80 D S y 60 y 300 3 2 40 3 200 420 100 y 1 A 4 0 o ty ty t3 ta ty te ty 11 30AM 12 00PM o
4. off i i i L L m Figure 18 Acoustic pump status detection in water flow line real world near full blockage in oil flowlines 4 10 Evaluation Summary and Algorithm Generalization We have shown our multi modal approach works well to detect cold oil and hot water blockages We next consider related problems and possible future work We believe our general approach use of two different types of sensors one accurate but noisy and the second to filter the noise generalizes to other sensing problems Our algorithm may apply to other applications where blockages occur in addition to cold oil lines For example vehicle or building cooling systems may have similar blockage prob lems in systems that duty cycle as do hot water distribution systems These systems both have fluid moving through a pipe network where flow can be detected by temperature variation from ambient yet one must also coordinate with driving machinery that operates intermittently and would otherwise cause incorrect outage detection Our short term field experiment for cold oil blockage was successful but additional work remains A next step is longer term testing to evaluate the robustness of the hard ware system and the tolerance of the detection algorithm against environment change Second a full integration with field network is a stepping stone to transform this research into a practical field equipment Third we are actively seek ing o
5. 00PM 6 00PM 7 00PM x 1 L 5 J500 B LA Ji time Figure 17 Temperature flow presence detection in water flowline ages at 3 00pm 4 00pm and 6 30pm when the the reference value k is set at the mid point of anomaly u and qual ity uo levels This delay occurs because learning u out of pump off temperature underestimates the temperature be havior at near full blockage The different temperature be havior is manifested by the two facts temperature drop at in op is more gradual than pump off drop and temperature rebounce a little at non op In another words the suboptimal u makes the algorithm insensitive to the temperature drops because it tunes k too low and the lower k the longer before temperature drops below k and triggers detection There fore to adjust the parameters of our algorithm for near full blockage detection we increase u by 20 every time before computing a new k We next evaluate our acoustic pump status detection and find the overall accuracy is 100 shown in Figure I8 More specifically all nine pump on and two pump off periods are correctly detected Like temperature detection above we adjust the parame ters in acoustic detection because the signal from the water recirculation pump is different from the pumpjack in the field test The water pump on noise is a wide band signal from the motor and fluid flow and unlike the oil pumpjack there is no periodic cycle and bursty rod t
6. 5 3 Multi Modal Sensing Applications We use cold oil blockage detection as a case study to learn multi modal sensing in industrial monitoring Next we re view related multi modality work in the same domain fol lowed by a broader review in other application domains 5 3 1 Multi Modal Sensing in Industrial Monitoring Our cold oil blockage detection shows complementary collaboration with multi modal sensing has potentially great feasibility to industrial monitoring applications A few other multi modal sensing applications are targeted to the same field 90251 Zeng et al show using vibration force and acoustic emis sion sensors to monitor health of high speed milling machine and predict wear out 5I Our pumpjack status detection is similar to their idea rumbling machines emit detectable acoustic pattern However we are not doing oil pipe block age prediction and it is part of our future work Futagawa et al design an integrated electrical conductivity temperature sensor for cattle health monitoring 9 Their sensing is inva sive since they embed sensors into rumens of cattle while part of the requirement of our oil line blockage detection system is non invasive to lower cost Gupta et al use mi crowave and eddy current image to evaluate corrosion under aircraft paint and in lap joints 12 They collaborate two modalities competitively meaning either on can fulfill the task but fusion achieves better results but our sensor colla
7. T We further visualize one data set T as an example to better demonstrate that this claim Figure 4 clearly shows that the data by mote is merely off from ground truth by a constant but the fluctuation is almost the same For clearer comparison we post facto ly convert raw ADC reading by mote to Celsius scale under following equation 53 ADC x 3 x 1000 2 852 where B is 367 as the gain of our pre amplifier board In our detection task the algorithm is more sensitive to temperature drops instead of the absolute value and hence a constant dis parity is acceptable 100 1 mote 80 USB O L e ne 6OF en erm pratt eg PN na eat E Pe S SPN i a Sot 2 a ee aad ie wa 2 40 paat g 2 20 KY J 0 L L 1 i i i 11 00AM 12 00PM_ 1 00PM_ 2 00PM_ 3 00PM_ 4 00PM_ 5 00PM time Figure 4 Temperature measured by mote and USB data logger at T2 4 1 2 Acoustic Sensor Measurement We first look into our acoustic mote Before we com pare mote and PC acoustic sensors we briefly describes their components and the difference in the data collection approaches Our acoustic mote is consist of Mica 2 mote an electret condenser microphone and a Mica sensor board Whereas our PC acoustic system is equipped with more powerful hardware a laptop with sound card complying to Intel high definition audio architecture and a battery powered lavalier microphone The hardware superiority of PC system alone
8. accidentally updating threshold to an inappropriate value 2 4 Acoustic Sensing to Avoid False Alarms Our discussion in Section shows that temperature alone is not enough Acoustic sensing on pumpjack status can avoid the false alarms caused by regular pump off In this section we describe our acoustic algorithm design and next discuss how we automatically tune parameters in that algorithm We need to determine if pumpjack is operating for end pipe blockage detection Since pumpjack stroke with engine rumbling generates wide band noise and propagates along pipe we use microphone mounted on pipe surface to mea sure the sound pressure level SPL a high level of which suggests pumpjack operating When pumpjack is off micro phones are expected to pick up much lower energy of envi ronmental noise Our acoustic algorithm works as follows First for each stroke cycle C we detect if pumpjack is on by comparing sound amplitude to a pre configured threshold 0 If sam ples in C exceeds the threshold mostly because of a signifi cantly loud rod tube clang noise associated with each stroke we decide the pumpjack is on during the whole cycle typ ically 7s However simple pumpjack flip detection is not robust against transient error and hence we need to know if the pumpjack is steady on In order to make that decision we check a longer history to see if it was being on for a whole warm up period W long usually far longer than a single cy
9. acoustic detection system acoustic 0 1 0 08 pu 3 s s 2 0 06 z a og Ey o o o 0 04 A a 2 D z Bi 0 02 __ E w 11 00AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM on x i X a Xx J 1 1 n offf i i 1 Ut 1 L 1 1 1 L L time Figure 11 Acoustic pumpjack status detection result by PC Table 2 The accuracies of blockage detection correct incorrect location tp tn fp fn Accugy Ty 8 2 0 80 T 3 6 1 0 90 T 3 6 1 0 90 total event 10 performs reasonably well and it is easy for our algorithm to achieve perfect accuracy on a cleaner dataset We next eval uate the end blockage detection on top of both temperature and acoustic sensing 4 7 Blockage Detection Accuracy In order to review our end blockage algorithm perfor mance we deploy a full system in our field test The block age detection fuses both pumpjack status and flow presence to determine if pipe is clogged or not In the section we evaluate its accuracy after defining the metrics below Like temperature and acoustic sensing Section 4 3 we define the metrics of blockage detection in a similar event based manner Since blockage detection is mostly based on temperature sensing event is likewise defined by experiment interval The terms to denote the correctness is as follows a True Positive is when the algorithm correctly declares an emulated blo
10. cle Hence to correctly detect pumpjack status we need to properly configure three parameters certainty of drop C warm up period W and threshold 6 We do tuning on base station because the training involves certain intensive computation as auto correlation and memory storage com plexity both beyond mote capacity so we employ a PC in our experiments In principle a mobile phone class proces sor could easily accommodate this work although it is be yond 8 bit motes The on site training step makes acous tic sensors robust against environment noise and mechanical difference across pumpjacks We next describe our training algorithm for these three parameters started by training data collection Before deployment we collect a short period of acoustic training data containing both pump on and off We next compute C by running auto correlation over the pump on trace The lag yielding the largest coefficient represents pumpstroke cycle To prevent from choosing harmonics in implementation we search the highest coefficient in a possible cycle range say 5s 9s Further based on our prior study W could be set as five times of C We consider both pump on and off to compute 0p be cause it needs to be able to properly denote the difference between those two status We first compute the noise floor by averaging all the samples in pump off trace We next throw away all samples below noise floor in pump on seg mentation 0 equa
11. easily deployed to address many real world problems from sewage pipe leakage detec tion B7 milling machine wear out prediction and live stock health monitoring 9 In spite of their effectiveness in some applications sen sornet uptake has been slow in many industrial applications SCADA systems today often employ traditional dedicated and often expensive sensors or fall back on manual observa tions where automated sensing is not seen as cost effective A challenge in use of low cost wireless sensors is that sim ple sensing methods often create many false alarms when they are confused by noise or changes in regular operation In this paper we propose to use different kinds of sensors to distinguish real anomalies from false alarms We select a main sensor that detects the anomaly but may be confused by changes during regular operation We then add additional sensors that can distinguish actual problems from false posi tives although they cannot detect anomalies alone This research is partially supported by CiSoft Center for In teractive Smart Oilfield Technologies a Center of Research Ex cellence and Academic Training and a joint venture between the University of Southern California and Chevron Corporation t Revisions in July include correction of typos and small clar ifications in Section 2 1 2 3 2 4 4 2 4 4 4 6 and John Heidemann Information Sciences Institute University of Southern California Marina de
12. in tervals to allow the system time to stabilize between changes Table I shows our schedule with three pump off periods for all four temperature motes to learn u and update k with the last one 28 minute long ran shorter than the first two each about 50 minute long due to time constraint According to the blockage introduction in Section in reality we may generally categorize blockage in to two types regarding how it is formed One is caused by a lump of viscous oil or sand clogging narrow fitting during pumpjack operation op an in op blockage The other is caused by residue oil in pipe cooling off and turning solid during shut in before pumpjack resumes operation a non op blockage To better evaluate the generality of our algorithm we simulate both types in three instances over the course of the day and each stage runs between 24 to 45 minutes The simulations are interleaved with other two types of stages One is valve open and pipe temperature rebounce so sensors can learn normal pipe tem perature uo during operation The other is pumpjack shut in which configures the sensors CUSUM anomaly level py i e temperature on stagnant flow In addition to this field test we carry out two prior field experiments where we evaluate components of our system and collect ground truth data for analysis in the lab Prior tests were done at a different wellhead We omit this data here due to space but replay of this ground truth data in th
13. rates with low cost sensors While some prior sensors have explored multi modal sensing with expensive sensors for example cameras 10 and PC level computation including mobile phones or laptops BISI we believe we are the first to show these approaches apply to low cost embedded sensors Our second contribution is to prove this claim by explor ing a specific application we design an embedded sensing approach that detects cold oil line blockages using a combi nation of inexpensive temperature and acoustic sensors Sec tion B then test our specific implementation Section B in the field Section 2 Design of Cold Oil Blockage Detection Al gorithm Here we define the problem we are solving then explore how low cost temperature sensors detect blockage acous tic sensors detect equipment operation and the two together provide reliable blockage detection with a low false positive rate 2 1 Problem Statement The goal of this paper is to understand how sensing can assist industrial applications and how multi modal sensing can help avoid false positives While in the abstract multi modal sensing is straightforward the key question is under standing how real world sources of noise and false detec tions affect sensing system design To that end we focus on cold oil blockage as a real world application The Problem Cold oil blockage occurs when the return line from a producing oil well becomes blocked typically due to change
14. tion temperature lower than the pipe skin temperature in op eration Although it is not mentioned in its specification 46 we believe this model on Mica sensor board WM 62A is not designed for a higher temperature task Even if it sus tains the heat electret condenser microphone has an unpre dictable frequency response under high temperature around 80 C 49 Therefore in deployment we apply thermal in sulation on top of the over warm pipe to protect our micro phone The insulation is called Fire Blanket and is made of woven fiberglass We are aware of some side effects of sand wiching insulation between the microphone and pipe for ex ample signal attenuation However under the design prin ciple of low cost sensing we decide to make this trade off instead of employing expensive specially customized micro phones say US 5 000 priced Briiel amp Kjzr 4949 automotive surface microphone Finally the design principle of temperature motes inher its our prior work 52 They each consists of a Mica 2 for control a custom amplifier board to optimize thermocouple signal readings and a thermocouple sensor NANMAC D6 60 J J type for pipe line and ambient temperature measure ments Figure 3 c and 3 shows how we deploy them in our experiment During experiments we were surprised to find that our custom amplifier boards are sensitive to their operation tem perature although all components are rated at a much higher range O
15. 05 2 s ty 100 er er a a 0 k ToO POE E a 9 re o i i normalized product 95 1 1 1 1 1 1 1 1 1 1 JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC 100 50 below baseline 0 Figure 1 Seasonality analysis shows winter production loss rizes for each month how often that month s production is below its index It further shows that winter months Novem ber through February witness consistently lower production rates There are multiple factors that contribute to this trend scheduled maintenance is often planed to avoid hot summer months but field engineers confirm that a significant factor to reduced winter production are well problems due to cold oil blockage Although we focus on cold oil blockage so can evaluate real world sources of error in Section 4 10 we consider how multi modal sensing applies in other applications Current Approaches Aware of the problem oil com panies have explored current sensing approaches including flowmeters pressure sensors leak detection and of course manual inspection Unfortunately currently techniques for automation have high installation and maintenance costs For example pressure sensors cost US 1 000 or more to pur chase and have annual costs of 300 or more to recalibrate flow sensors are more costly As a result these sensors are used only on a few very productive wells while manual in spection remains common in spite of it
16. 44 In Ad Hoc Mobile and Wireless Networks volume 2865 of LNCS pages 223 234 Springer Berlin Heidelberg 2003 S Moncrieff S Venkatesh and G West Dynamic privacy assessment in a smart house environment using multimodal sensing ACM Trans Multimedia Comput Commun Appl 5 10 1 10 29 Nov 2008 N Oliver and E Horvitz S seer Selective perception in a multimodal office activity recognition system In S Bengio and H Bourlard editors Machine Learning for Multimodal Interaction volume 3361 of Lecture Notes in Computer Sci ence pages 122 135 Springer Berlin Heidelberg 2005 E S Page Continuous Inspection Schemes 41 1 2 100 115 1954 Y Qu T Wang and Z Zhu An active multimodal sensing platform for remote voice detection In Advanced Intelligent Mechatronics AIM 2010 IEEE ASME International Confer ence on pages 627 632 july 2010 C J V Rijsbergen Information Retrieval Butterworth Heinemann Newton MA USA 2nd edition 1979 A Singhal and C Brown Dynamic bayes net approach to multimodal sensor fusion In Proceedings of the SPIE The International Society for Optical Engineering pages 2 10 1997 D Sinha Acoustic sensor for pipeline monitoring Tech nical Report LA UR 05 6025 Los Alamos National Labora tory July 2005 T Stiefmeier D Roggen G Troster G Ogris and P Lukow icz Wearable activity tracking in car manufacturing Perva sive Computing IEEE 7 2 42 50 a
17. 57 MMUI 05 pages 25 32 Sydney Australia 2006 Australian Computer Society Inc 12 K Gupta M Ghasr S Kharkovsky R Zoughi R Stanley A Padwal M O Keefe D Palmer J Blackshire G Steffes and N Wood Fusion of microwave and eddy current data for a multi modal approach in evaluating corrosion under paint and in lap joints Review of Quantitative Nondestructive Eval uation 26 611 618 2007 13 K Harrington and H Siegelmann Adaptive multi modal sen sors In M Lungarella F Iida J Bongard and R Pfeifer editors 50 Years of Artificial Intelligence volume 4850 of Lecture Notes in Computer Science pages 164 173 Springer Berlin Heidelberg 2007 14 W Hu N Bulusu C T Chou S Jha A Taylor and V N Tran Design and evaluation of a hybrid sensor network for cane toad monitoring ACM Transactions on Sensor Net works 5 1 4 1 4 28 Feb 2009 15 P S Huang X Zhuang and M Hasegawa Johnson Improv 16 17 18 19 20 21 22 23 24 25 26 27 28 ing acoustic event detection using generalizable visual fea tures and multi modality modeling In Acoustics Speech and Signal Processing ICASSP 2011 IEEE International Con ference on pages 349 352 may 2011 V Jacobson Congestion avoidance and control In Sympo sium proceedings on Communications Architectures and Pro tocols SIGCOMM 88 pages 314 329 Stanford Califo
18. 7 1222 Aug 2006 C Zhang A Syed Y H Cho and J Heidemann Steam powered sensing In Proceedings of the 9th ACM Conference on Embedded Networked Sensor Systems SenSys 11 pages 204 217 Seattle Washington USA Nov 2011 ACM C Zhang A Syed Y H Cho and J Heidemann Steam powered sensing Extended design and evaluation Techni cal Report ISI TR 2011 670 USC Information Sciences In stitute Feb 2011 F Zhao J Shin and J Reich Information driven dynamic sensor collaboration for tracking applications JEEE Signal Processing Magazine 19 2 Mar 2002 J Zhu Application of temperature method for jam detection in BF coal injection Baosteel Technology 6 15 18 2005 X Zou and B Bhanu Tracking humans using multi modal fusion In JEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops page 4 june 2005
19. G Tzanetakis Multimodal sensor analysis of sitar performance Where is the beat In Multimedia Signal Processing 2007 MMSP 2007 IEEE 9th Workshop on pages 74 77 Oct 2007 7 Y Chamber A Garg V Mithal I Brugere M Lau V Kr ishna S Boriah M Steinbach V Kumar C Potter and S Klooster A novel time series based approach to detect gradual vegetation changes in forests In Proceedings of the 2011 NASA Conference on Intelligent Data Understanding pages 248 262 Mountainview CA USA Oct 2011 NASA 8 A Daneels and W Salter What is SCADA In Proceedings of the International Conference on Accelerator and Large Ex perimental Physics Control System ICALEPCS 99 Trieste Italy Oct 1999 9 M Futagawa M Ishida M Ishida and K Sawada Study of a wireless multimodal sensing system integrated with an electrical conductivity sensor and a temperature sensor for the health control of cows IEEE Transactions on Electrical and Electronic Engineering 6 2 93 96 2011 10 L Girod and D Estrin Robust range estimation using acous tic and multimodal sensing In Proceedings of the IEEE RSJ International Conference on Intelligent Robots and Systems IROS Maui Hawaii USA Oct 2001 IEEE 11 R Goecke Audio video automatic speech recognition an example of improved performance through multimodal sen sor input In Proceedings of the 2005 NICTA HCSNet Mul timodal User Interaction Workshop Volume
20. Reducing False Alarms with Multi modal Sensing for Pipeline Blockage Extended ISI Technical Report ISI TR 2013 686b June 2013 revised 3 July 2013 Chengjie Zhang Information Sciences Institute University of Southern California Marina del Rey California USA chengjie isi edu Abstract Industrial sensing applications place a premium on cost effectiveness and accuracy Traditional approaches often use expensive invasive sensors because inexpensive sensors suffer from false positive detections Sensor cost means au tomation is sparse or avoided when the value of specific sites cannot be justified In this paper we show combining dif ferent types of sensors can allow low cost sensors to avoid false positives enable much greater levels of automation in some applications We explore this problem by studying a specific application blockages in oil flowline common in cold weather We use pipe skin temperature to infer changes in fluid flow and combine readings with acoustic data to avoid false positives and be robust to environmental changes We demonstrate that our approach is effective with field ex periments Finally suggest that this approach generalizes to other classes of problems where false positives from one sensing modality can be resolved by multi modal sensing 1 Introduction Sensor networks are used to collect data detect prob lems and take actions in the physical world Small and inexpensive sensornets can be
21. S NED ata sheets 2006 http www audio technica com cms wired_mics 9c6ecal7168eef6f index html ATR3350 omnidirec tional condenser lavalier microphone 45 46 47 48 49 50 51 52 53 54 55 56 http www dataq com support documentation pdf datasheets el usb tc data logger pdf EL USB TC thermocouple data logger datasheet 2012 http www panasonic com industrial components pdf em05_wm62_a_c_cc_k_b_dne pdf WM 62A datasheet http www vernier com go gotemp html Vernier Go Temp thermometer X Wang H Qi and S Iyengar Collaborative multi modality target classification in distributed sensor networks In Infor mation Fusion 2002 Proceedings of the Fifth International Conference on volume 1 pages 285 290 2002 Y Yasuno and J Ohga Temperature characteristics of electret condenser microphones In Electrets 2005 ISE 12 2005 12th International Symposium on pages 412 415 Sept 2005 S Yoon W Ye J Heidemann B Littlefield and C Shahabi SWATS Wireless sensor networks for steamflood and water flood pipeline monitoring IEEE Network Magazine 2009 accepted in 2009 to appear in 2010 or later H Zeng T B Thoe X Li and J Zhou Multi modal sensing for machine health monitoring in high speed machining In Industrial Informatics 2006 IEEE International Conference on pages 121
22. alve means the production valve on T accuracy We study the hardware difference helping PC mi crophone collect better data with higher SNR for future mote improvement guidance We focus on the three most impor tant stages in line with converting raw vibration to digital sig nal microphone preamplifier and analog digital converter ADC After comparing the specifications we find that PC beats mote microphone in all the three stages First although mote microphones have higher sensitivity 45 dB 46 than the one with PC 54dB 44 but PC microphone has a much smaller resistance 1 kQ against 2 2kQ potentially able to produce larger current under the same sound pres sure Second our PC has better preamplifier embedded in its sound card SoundMAX AD1988A than mote does 42 For example the former has higher input impedance but lower total harmonic distortion Finally the ADC on the microcontroller of our motes have a much lower resolution 10 bit comparing to PC sound card s 24 bit resolution In short all these differences results in better quality of PC data with larger SNR than the ones collected by our acoustic mote system In this section we evaluated our acoustic mote for pump jack status detection We further applied the same algorithm to PC acoustic data to provide more complete verification of our acoustic detection approach Based on experiment re sults we conclude that our mote
23. ause is parameter mis configuration Due to the im perfect pumpjack detection an overall accuracy of 87 in Section reported pumpjack status often incorrectly flips in the middle of an event causing mis configuration on anomaly and quality levels However surprisingly a rela tive chaotic parameter auto configuration scheme does not throw off our entire blockage detection Our algorithm ex hibits robustness against the configuration errors which at best cause only one false positive on T alone after 1 09pm in Figure I2 a p In all our blockage detection algorithm has a high ac curacy 80 in the worst case A close look on algorithm output plots shows what causes mis detection However the interesting results raises one further question about the ro bustness of our algorithm against the jittering in pumpjack detection We next investigate this issue in Section 4 8 Robustness of Blockage Detection The results in evaluating our blockage detection surprise us because blockage detection is often insensitive to the error in pumpjack status detection after fusing it with flow pres ence detection One commonality in these errors is that the threshold to detect pump on is such mis configured that de tection frequently in minutes flips between pump on and off In order to verify if this feature is systematic or random a 400 pump
24. b oration is complementary 5 3 2 Multi Modal Sensing in Academic Projects Many academic studies explored multi modal sensing for different applications including general sensor fusion 20 target classification 48 target tracking 56 human activity or health monitoring 3 6 15 29 30 36 Robotic navigation and human computer multi interaction 11 27 321 38 Some work uses multivariate statistics modeling for mul tiple sensor data fusion Goecke uses coinertia analysis to find a mathematical compromise between the correlation of audio and video 3D lip tracking features 11 Kushwaha et al uses separate non parametric model for audio sensors and parametric mixture of gaussian model for video sensor in their vehicle tracking application 22 Annavaram et al use bivariate model for ECG and accelerometer data to monitor sensor bearer s activity 3 Our oil line blockage detection algorithm has a similar way to do change point detection We build separate models for both acoustic and temperature data and configure thresholds for each In addition we plan leverage the correlation between the two modality when pipe is normal Other work specifically uses Bayesian networks Tamura et al use triphone HMM to model audio video data for speech recognition They find that a training set of audio visual data achieves better recognition accuracy than audio only data 38 McGuire et al leverage Bayesian networks to integrate spoke
25. ckage a True Negative is when flow is normal or the pump is off and the algorithm remains silent a False Positive is when flow is normal or the pump is off but the algorithm incorrectly declares a blockage a False Negative is when the algorithm mis detects an emulated blockage We then accordingly defines the overall accuracy Accu We first evaluate Accu after fusing both temperature and acoustic results Tabet shows that overall accuracy of our fully automated system is between 80 and 90 This result further shows our blockage detection algorithm is very accurate This table suggests two further observations First all temperature drops caused by blockages are correctly de tected because no false negative occurs across all three sites This sensitivity of our blockage detection algorithm to tem perature drop is consistent with the result we have in eval uating our flow presence detection in Section 4 4 In addi tion no false positive further indicates that our algorithm is general to different situations because we emulate different blockages which forms either during pump shut in or during pump operation The second observation is the detection period is short meaning our system is able to give rapid feedback In prob lem statement Section 2 1 we explain why rapid feedback is important to mitigate the loss We find it generally takes between 10 to 30 minutes before our algorithm triggers The third observation is that so
26. ct pump jack operation And one acoustic sensor placed near the wellhead can provide pumpjack status for all temperature sensors on the production line temperatures on the same line downstream The acoustic sensor listens to flow in the pipe and the clanging of the pumpjack rods and tubing to detect pumpjack operation These sounds propagate well through the pipe so acoustic sensor placement can be within 20 m of the wellhead Our hypothesis is that our combination of temperature and acoustic sensing is both necessary and sufficient to detect cold oil blockage Sources of noise Although we focus on pumpjack op eration as our main source of error we must consider many sources of noise from the environment field and measure ment system Environmental noise contains diurnal and seasonal changes in weather and ambient temperature Pipe skin tem perature changes by a few C over the course of a day due to changes in sunlight and wind or other weather Our algo rithm is insensitive to this change because the temperature difference between normal flow and ambient is much larger Seasonal weather changes have a greater change with tem peratures that vary by 38 C or more from the min in winter to the max in summer However this long term change does not affect our algorithm because the detection threshold is hourly auto retrained against recent pipe skin temperature quickly adapting our algorithm in as short as hours Second fiel
27. d conditions change including downhole conditions equipment maintenance and main line back pres sure Downhole temperature and pressure changes as the field produces oil and due to changes in injection These changes are generally slow over days or weeks our algo rithm retrains hourly and so adapts to these Valve close up caused by maintenance indeed behaves similar to a real sud den full blockage We depend on field engineers to identify maintenance a possible source of false blockage detections Finally there will be some temperature propagation from the main line back to a blockage We expect this effect to be minimal The last group is measurement noise which is related to our deployment setting including sensor installation and random glitch If a sensor has a loose contact with the pipe the readings are always a weighted average between ambi ent and pipe skin temperature Poor connection will reduce our algorithm s sensitivity but our tuning accounts for varia tions We confirm in tests that our algorithm adapts to loose connections that cut the mid point between ambient and nor mal pipe operation temperature in half still finding the cor rect reference value and triggers on sub 20 C drop From the discussion of three categories of noise environmental system and measurement we conclude that our algorithm with parameter auto tuning is robust enough 2 3 Temperature Sensing for Flow Presence Section 2 2 shows flo
28. decrease case above depicted in Figure Summarizing how it response to all the three models we conclude that our blockage algorithm is systematically ro bust against the negative effect of jittering pumpjack status detection on parameter setting The most significant reason is that under those circumstances the reference value cor rectly stay above or below the temperature by tracking it as low pass filtered signal ki z x Iwg dung P AS Water a Logical view of lab test b Physical view of lab test Figure 16 In lab near full blockage test on water 4 9 In lab Near Full Blockage Detection In prior sections we show that our multi modal detection correctly detects full blockages in the field However full blockages can quickly result in damage to equipment so we would like to detect blockages before they fully close the line We therefore next extend our work to detect near full blockage We verify this extension with laboratory tests Un like full blockage we do not evaluate near full blockage in oil field because it is not safe to emulate a realistic one opening production valve slightly but with circulation valve closed would cause over high pressure at the wellhead In stead we use a mock up flowline system in the laboratory We choose hot water as the fluid because it has some similar properties as oil both are incompressible and moderately warmer than ambi
29. e lab shows our system works correctly on another well with different sensor locations 4 3 Evaluation Metrics Section 2 shows that we detects cold oil blockage by fil tering out irrelevant flow absence with acoustic pumpjack status detection Before evaluation we describe below the temperature and acoustic detection metrics due to their sim ilarity We evaluate both temperature and acoustic sensing in an event based manner but with separate event definition For temperature one event is one interval between changes of equipment setting because we care about if flow presence Production valve Circulation valve uoryepnoma oul uononpoig Reservoir a Logical view of deployment E 5 i te tne lt i G Circulat valve b Physical view of deployment Figure 6 November 2012 field deployment detection triggers or does not trigger eventually in certain conditions Each event starts with setting change including valve close up open and pump on off and ends with another change retaining the same setting across the entire event According to the schedule in Table I we divide our experi ment after 11 11am into ten stand alone events We discard events less than ten minutes the period between 11 01am and 11 llam because our algorithm requires 15 minutes to stabilize and our algorithm is still learning parameters Hence we first define metrics for temperature det
30. e false positives can be re solved by a different sensing modality Acknowledgments We would like to thank Andrew Goodney for his input on acoustic sensing We thank Greg LaFramboise Charlie Webb Mohammad Heidari for their input on Chevron s business requirements and their assistance in our field experiments Iraj Erhagi and Mike Hauser for their guidance as co directors of CiSoft 7 References 1 V V A Tartakovsky Change point detection in multichannel and distributed systems with applications In Applications of Sequential Methodologies pages 331 363 2004 2 B R Abidi N R Aragam Y Yao and M A Abidi Survey and analysis of multimodal sensor planning and integration for wide area surveillance ACM Computer Survey 41 7 1 7 36 Jan 2009 3 M Annavaram N Medvidovic U Mitra S Narayanan G Sukhatme Z Meng S Qiu R Kumar G Thatte and D Spruijt Metz Multimodal sensing for pediatric obesity applications In Proceedings of the International Workshop on Urban Community and Social Applications of Networked Sensing Systems UrbanSense Raleigh NC USA Nov 2008 ACM 4 E I Barakova and T Lourens Event based self supervised temporal integration for multimodal sensor data Journal of Integrative Neuroscience 4 2 265 282 2005 5 M Basseville and I V Nikiforov Detection of abrupt changes theory and application Prentice Hall Inc 1993 6 M Benning A Kapur B Till and
31. e second Table 3 In lab Experiment schedule Artery start pump valve purpose 1 00pm on T learns uo 1 15pm off T learns py 1 56pm 90 off non op 2 29pm on open T learns uo 2 50pm in op 3 31pm off 90 off T learns u _ 4 04pm non op 4 35pm open T learns uo 5 15pm on 90 off in op 5 55pm open T learns uo 6 25pm 90 off in op long samples measuring average noise amplitude We later run our detection algorithm over these aggregated samples Finally we do not use fieldable hardware in our lab near full blockage test contrary to our field test for full blockage The hardware difference is because the field test hardware are designed and assembled for oil field environment and hence it takes extra amount of work on tuning them for the lab environment Fortunately the key properties of the two different fluids oil gas water mix in the field network and water in our laboratory network are similar enough that the success of this lab test demonstrates that our multi modal sensing can detect near full blockages in oil field In addi tion this lab test generalizes our algorithm to applications other than those in oil industry Our laboratory tests follow the same procedure as our field tests Section 4 2 however here we emulate near full blockage by closing about 90 the valve rather than clos ing it complete Following Figure we close valve V a on t
32. ecessary because if a uniform reference value were mis configured above 229 sensor downstream to production valve T3 would start to trigger false positives pump off i 7 m i 400 oO 4 300 3001 1 1 K booo g 1 t i 1 1 i Oo Li i 1 lt Hi 1 1 1 I 1 1 y lt 200 i a A otf i lt lt A oa lt lt 200 200 g E BAJE E E EE 2 2 ig ig STS WE es 2 i lt ids Zia I lt lt ka ig a A 2 D S Sa w EY ais A 468 6 8 6 8 2 E ab fa DE 10i sis 2 e g 2 ig ig lg 10053 ais 8 Ani E OE 1 18 E ok EWS aE of a WE 88 o O o o 2 me gS g g gg 8 BE 3 fo 2 fF fe a ee g g MEE 8 88 Or ODER LOMPAT i i R 8 PR m e iF aim g g 8 gf OB IS LEUOAM 12 00PM SE OUE A 2 00PM 3 00PM AO0EM S00PM 11 80AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM 11 80AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM lt gt x X X 4 XxX K X x 30007 Pa rd Po oy a 3000777 Am ae 30007 i a we ot ft dt 4 i of a eee o f rt eee ee Eo hi 1 i 1 T o nm E i He v Hl 1 fi hi i o i 1 ai bo J i i ron an a h l i di a oi E l l j oad a Upstream T time time b Downstream before production valve T c Downstream after production valve cr Figure 7 Flow presence detection results 4 6 Detecting Pumpjack Operation Previous evaluation shows flow pr
33. ection to denote the correctness of flow presence for each event a True Positive tp is when flow stops due to either pump is off or a valve in line is closed and the algorithm triggers during the whole event a True Negative tn is when flow is normal and the algorithm does not triggers anytime during the event a False Positive fp is when flow is normal but the algorithm incorrectly triggers and a False Negative fn is when flow stops but the algorithm incorrectly keeps silent And we define overall accuracy using terms from informa tion retrieval B3 tp in tp in fp jn Contrary to temperature the event in acoustic evaluation is defined by a sample i e one second long sensor reading because we care about instantaneous pumpjack detection We use similar ways to define the four terms tp tn fp and fn out of the pairwise combination between pump on off and algorithm output on off For example it is a tp if pumpjack is on and the algorithm correctly declares it on We inherit the same equation to compute overall accuracy as in forgoing temperature metrics We can then define accu racy of pump on and off events using subsets of these mea surements Accug A P ccu on tp fn tn A CCU off fp In our experiment our acoustic node log 23 129 valid sam ples Among them 15310 are tp 4893 are tn 2227 are fp and 699 are fn After we define metrics we next evaluate our tempera ture flow p
34. ent We next introduce our lab experiment and show the results of our multi modal sensing on near full blockage To evaluate partial blockaged detection we set up a testbed in our lab We constructed a recirculating network of hot water similar to that used in our prior work 53 It consists of a tankless water heater a recirculation pump a plastic lidless tank and a small network of PVC pipes and valves Figure 16 Because we are evaluating multi modal sensing we deploy both temperature and acoustic sensors To detect pipe skin temperature we tape down USB based Go Temp temperature sensor on the artery line after a valve Figure 16 a Our acoustic sensor is a lavalier micro phone the same as the one used in our field test to collect ground truth data For acoustic detection we must account for differences between the signal of pumpjack operation in the field and the water recirculation pump in the lab The major differ ence between pump on and off is average amplitude and our algorithm is still able to distinguish the two when we adjust parameters Please see detailed explanation on signal difference and evaluation result in later this section In de ployment we tape down the microphone on the recirculation pump to detect the pump on off status by picking up pump operating noise Similar to the field test Section 4 1 we collect raw pump noise with a sampling rate of 8 kHz and down sample to 2 kHz before aggregating into on
35. esence detection by temperature is effective However inferring blockage solely on flow absence is not enough because regular pumpjack shut in too stops fluid flow details in Section 2 1 To avoid false alarms caused by these irrelevant temperature drop we use acoustic sensing to detect pumpjack status and later apply the result on top of temperature Next we evaluate acoustic algorithm accuracy and draw conclusions based on the results We first evaluate the overall pump on and off detec tion accuracies The overall accuracy is high 20203 out of 23 129 events defined by samples in Section 4 3 are correct 87 and the accuracy of detecting pump on is even higher 96 However the accuracy of detecting pump off is 69 which is low compared to the other two metrics Next we vi sualize the algorithm output trace to investigate why pump off detection only works partially Pump off detection is much less accurate than overall and pump on detection To understand the difference we need more information about why many pump off samples trig ger pump on detection Figure visualize the details by showing the acoustic amplification trace and algorithm out puts on mote A red thick line in the upper plot indicates the threshold 0 we used in our field test The high spikes fol low every valve status change is because the relatively loud noise generated by wrench valve clanging is captured by the acoustic sensor One reason pump off detec
36. geneous sensor networks In Proceedings of the 6th international conference on Informa tion processing in sensor networks IPSN 07 pages 519 528 Cambridge Massachusetts USA 2007 ACM E Liu C Li S Peng and X Wu Detection technology for oil pipeline plugging based on decompression wave method In Computational and Information Sciences ICCIS 2010 In ternational Conference on pages 820 823 Dec 2010 L Liu and S L Scott A new method to locate partial block ages in subsea flowlines In SPE Annual Technical Conference and Exhibition New Orleans Louisiana USA Sept 2001 Society of Petroleum Engineers SPE 71548 P Martin Z Charbiwala and M Srivastava Doubledip leveraging thermoelectric harvesting for low power moni toring of sporadic water use In Proceedings of the 10th ACM Conference on Embedded Network Sensor Systems SenSys 12 pages 225 238 Toronto Ontario Canada 2012 ACM P McGuire J Fritsch J Steil F Rothling G Fink S Wachsmuth G Sagerer and H Ritter Multi modal human machine communication for instructing robot grasp ing tasks In Intelligent Robots and Systems 2002 IEEE RSJ International Conference on volume 2 pages 1082 1088 2002 F Molina J Barbancho and J Luque Automated meter reading and SCADA application for wireless se nsor network 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43
37. he artery line in and open the branch valve to keep water flow and the heater running Table 3 shows our lab test schedule with two pump off periods and five near full blockages In the six hour long test we collect both temper ature and acoustic traces by PC We run our detection algo rithm on a PC and do analysis of the data after collection In principle we can integrate our algorithm with the sensors and run with exactly the same hardware and similar software to the field however here our goal is to test the generality of the algorithm so we do analysis off line to allow us to study a range of algorithm parameters post facto We use the same evaluation metrics as in field tests Section 4 3 Figure shows that our flow presence detection algo rithm with adjusted parameters gives rapid and accurate re sponse on abnormal flow pump off and near full blockage Certainty of drop Cz correctly builds up on all seven ab normal flow events within 20 minutes the lower plot of Fig ure 17 In addition no false positive is raised when we open up the valve to learn uo To get these results we must adapt the parameters to de tect near full blockages of the flow We found that the base algorithm takes too long about an hour to trigger on block 50 r temp 2 pump off pump off k v I i A 5 40F g fe a lt lt By ge gt JR g S S 8 5 og Sg 8 sig 8 S 515 R S X RR B iR 8 iR 1 00PM 2 00PM 3 00PM 4 00PM 5
38. he results from the three temperature sensors we make four observations First our approach achieves repeatable detection results because the results are consistent across all three sensors Second one sensor is enough to cover a large pipe segment for detection because it can detect blockage upstream and downstream to it Temper ature traces before and after the valve T and T is almost equivalent shown by a high positive correlation 0 98 and low standard deviation in differential 1 5 C Therefore in our case either of them is enough for the pipe section after production circulation branch at least 20 m long to the next junction Three our sensor placement shows minimal recir culation from the main line since the temperature falls even downstream of the blockage both T and T show similar temperatures Finally the different responses upon block age between T and the other two shows multiple sensors can be used to locate blockages by distinguishing pipe locations with and without flow In this section we conclude that flow presence de tection with auto configuration achieves perfect result in field tests Therefore we next investigate the necessity of auto configuration by cross comparing between the three datasets 4 5 Auto Configuration of Temperature Mea surement In the previous review we demonstrate that our flow pres ence detection is accurate with parameter auto configuration on quality and anomaly levels We
39. his paper demonstrates a field tested system and evaluates specific sensing algorithms with multi modality 5 2 Change Point Detection Algorithms Many real time monitoring systems use abrupt detec tion or change point detection to detect problems in observed data In change point detection on flow pres ence we focus on cumulative sum control chart CUSUM and exponential weighted moving average EWMA for lightweight implementations on mote class platforms Our flow presence detection algorithms are inspired by several prior works 7 39 Some researchers use CUSUM to analyze time series and we implement it to detect flow presence Chamber et al develop a CUSUM based algorithm to detect vegetation changes in forests 7 Similar to their work we choose CUSUM for its capability in identifying small and gradual change and we too make our algorithm adaptive to noise They post process year long time series for small change while we in node process streaming data Another differ ence is their algorithm is designed to identifying multiple decreasing segments in a series but our algorithm focuses on immediate decision Several sensornets build on the simple EWMA algorithm from TCP 16 Trifa et al develop an adaptive alarm call de tection system for yellow bellied marmots using EWMA to update threshold by noise distribution estimation 39 Our work uses similar concepts to detect significant change in oil pipe temperature using EWMA
40. integrate several sensors into one wearable sensors to monitor worker s ac tivity They propose use one kind of sensing result to do automatic data segmentation for other sensing streams 36 Barakova and Lourens propose event based data fusion con trary to fixed time interval based fusion They use gyro scope data to segment visual data for robotic navigation 4 Our method employs this idea because we are actually using acoustic date to assist temperature date for blockage detec tion Our hypothesis indicates if pumpjack is not operating we have no way to tell if the pipe is blocked In other words we use acoustic date to segment temperature date and ignore those when pumpjack is off 6 Conclusion We have described a system for multi modal sensing to detect cold oil blockages We show combining different types of low cost non invasive sensors can avoid false posi tives and provide rapid blockage detection To achieve this we developed an algorithm first uses pipe skin temperature to infer changes in fluid flow for suggested blockages Later we suppress false alarms caused by some regular operation with acoustic sensing We have demonstrated the effectiveness of this algorithm and our implementation through field exper iments Although we have developed this system to solve cold oil blockage problem in oil field the principle of multi modal low cost sensor collaboration generalizes to other in dustrial sensing applications wher
41. is enough to justify its cleaner data Other than the hardware difference the second major dif ferences is sensor installation Although both sit on top of insulation we tape down mote microphone by duct tape to pipe while we clamp on PC microphone by a customized clasp and hose clamps likely to produce larger force to press the microphone against the pipe for a better contact Finally there is difference in sampling and aggregation mechanism in software after we abstracted out the OS differ ences although the final packet rates of both for evaluation are the same 1 Hz As we described in Section B the raw sampling rate on motes is 2 kHz and we next use a hierarchi cal aggregation to assemble one packet every second from ten 0 l second long short packets On the other hand the raw sampling rate on PC is 16 kHz much high than what we have with motes mainly because we plan to keep high quality ground truth data in case we need to investigate the frequency domain of acoustic signal Since the software we choose Audacity does not support a sampling rate as low as 1 Hz and more importantly we prefer to maintain the con sistency between both systems we re sample our PC data in gt f p nr amplitude Y L a 200 lt 100 11 00AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM Figure 5 Acoustic measured by mote and PC micro phone 2 kHz by the same software before we further aggregate it by one second
42. kin temperature and compares the result against pre configured thresholds We design our flow presence de tection with a similar idea However we use inexpensive and portable hardware US 500 while his centralized sys tem is likely expensive because one of its component is an industrial PC which usually costs US 1 500 Our prior work studies steam choke blockage detection with wireless sensor networks 52 We prove the feasibility of detecting steam choke blockage by non invasive temperature sensing We show similar sensing technique generalizes to a hot water distribution network similar to the oil retrieval line network in the paper In this prior work we develop a complete sys tem with thermal energy harvesting evaluated in a field test In this work all sensor motes are battery powered energy harvesting is our future work Unlike this prior work our current work adds multi modal sensing to address the false positives common to cold oil blockage In temperature sens ing the prior work takes differential temperature around the blockage point but here we use absolute temperature from one sensor because contrary to choke blockage the cold oil blockage location is unpredictable on a long line In addition our current work auto configures all parameters for detection during sensornet deployment We have previously explored the potential of sensor net works in oilfield production systems 50 While that work suggests the potential t
43. l Rey California USA johnh isi edu Our overall goal is to identify classes of industrial appli cations where multi modal sensing can resolve sensing am biguities In this paper we prove this claim in the context of a specific example cold oil blockages in flowlines in pro ducing oilfields A typical oilfield has many kilometers of distribution flowlines that collect crude oil extracted from wellhead pumpjacks gathers the oil for measurement and accounting and ultimately sends it to refineries Distribu tion systems near the wellhead are often small particularly in older fields In cold weather oil thickens because oil vis cosity has an inverse proportional relation with its tempera ture Oil may then interact with sand or other contaminants in the fluid and with pipe sags or narrow fittings resulting in blockages in the lines Blocks cause production loss and if left unresolved they can result in pipe leaks damage to the flowline or even to the pumpjack After pipe being fully closed it takes only tens of seconds for pressure to build up before some parts in line rupture Although the oil industry has explored several stand alone sensors current approaches are either unreliable or too ex pensive to install and maintain Section 2 1 Although some fields contain thousands of wells where production lines are vulnerable to blockage manual inspection is the most com monly used technique today Our insight is that multi modal se
44. ld up followed by blockage detection We evaluate the fusion result later in Section 3 System Implementation Before we review the details of our field experiment we briefly talk about the implementation of our mote sensing platform with low cost sensors We first briefly summarize the hardware of our multi modal sensing system Next we discuss the two challenges in acoustic node implementation and our software approaches to solve them 3 1 System Hardware Our sensor network consists of three types of nodes base node for data collection acoustic mote for pumpjack status detection and temperature mote for flow presence detection In this section we introduce the hardware of their parts Our base node is simply a Mica 2 mote connected to PC through MIB520 programming board It passively listens and logs all the packets transmitted from acoustic or temper ature motes in the network Our acoustic mote is composed of a Mica 2 mote and an MTS310CA Mica Sensor Board with an on board electret condenser microphone Panasonic WM 62A Figure B a p Figure 3 b shows how we tape and clamp the extended mi crophone on the pipe with thermal insulation and we discuss how decoupling microphone benefits signal gain later We cannot directly mount Mica 2 microphone on pipe because the high pipe temperature may damage the equip ment or at least result in inaccurate measurement Com mon electret condenser microphone has a sub 70 C opera
45. long window Despite the differences we listed above Figure 5 shows that our mote data is close enough to the ground truth It is difficult to directly convert their units hence we keep both trace in their raw units and hand scale them in the plot The correlation coefficient between the two traces is 0 44 prov ing a strong positive correlation The other observation that PC data has higher SNR which is depicted by much higher state transition spikes and near zero pump off noise detec tion In all through the comparison we show our mote data is close enough to PC data by a more expensive hardware suite This result further supports our hypothesis above that low cost sensor is capable of reaching effective yet economical sensing 4 2 Field Experiment Approach We next evaluate our system in field tests From 9 30am to 5 30pm November 7 2012 we evaluated our system and a producing oilfield in the California Central Valley working with field engineers from our research partners who operate that field During the nearly seven hour long exper iment our system collected acoustic and temperature traces did in node processing and ran the full detection algorithm We also collected ground truth data concurrent with opera tion of our experimental system Ground truth temperature and acoustic data employed USB thermocouple data loggers EL USB TC 45 and a laptop computer with a commodity microphone Figure 6 b shows the test site a
46. ls the 86 percentile of amplitude among all the rest of the pump on segmentation The reason we choose this value for O is that during a common 7s pump cycle our threshold should detect the single sample captur ing the loudest rod tube clang noise against other six under 1 Hz sampling rate Therefore the signature noise sample is likely to have a higher amplitude than the other 86 six out of seven in one cycle samples 2 5 Sensor Fusion for Blockage Detection We talk about the two algorithms in the above sections and next we describe how to fuse them to detect end block age If we interpret our basic hypothesis Section 2 1 with technical details we find that blockage could be detected as flow stops but pump is steady on In another words if pumpjack is off our algorithm ignores all suggested block age detection by temperature sensing although the certainty of drop builds up due to stagnant flow On the contrary if pumpjack is on our algorithm can de tect blockage all in the following two different situations If blockage occurs during pumpjack operation i e pumpjack is steady on we expect to witness a line temperature drop As soon as line temperature stays below reference value long enough blockage detection triggers Besides if blockage occurs during shut in after pipe cools off and pumpjack re sumes line temperature stays close to ambient and does not increase significantly Hence the certainty of drop can too bui
47. me false positives are raised We next evaluate why a perfect flow presence de tection does not lead to a perfect blockage detection To an swer this question we need to investigate the result on each event particularly on false positives The three figures in Figure 12 visualize our fully automated system outputs and show why false positives exist The lower plot in each fig ure contains both the certainty of drop C4 and the ultimate blockage indication We see that there is transients after the pumpjack resumes operation A blockage signal raised at 2 35pm in Figure 12 a because the pipe skin temperature resumes to normal sightly later than the temperature sensor first receives pump on signal We expect our base algorithm to remain silent although false positive is triggered How ever this can be easily fixed in an extended algorithm which suppresses anomaly outputs a short while after temperature sensor receives pump on signal Therefore we still count it a true negative in later evaluation One major cause of false positives on all three motes is incorrectly reporting pump on during the third pump off period 4 30pm 4 58pm under effective temperature drop detection However the block age detection algorithm successfully suppress the suggested blockages i e temperature drops in the first two pump off period because of a build in anti false alarm feature which ignores sporadic mis detection fp in pumpjack status The other c
48. n instruction visual memory and gesture based region bias to determine the object to be grasped by robotic arms 27 Zou and Bhanu evaluate both time delay neural network method and Bayesian network method for walking human detection from audio video data They conclude Bayesian network is better because of ease to train higher accuracy and clearer graphical model 56 Huang et al propose a coupled HMM method for audio visual joint modeling especially to solve asynchronization problem in a office activity monitoring application I5 Oliver and Horvitz use layered HMM a modular and hierarchical HMM method for office activity inference 30 Singhal and Brown uses Bayesian network to joint model audio and video data to predict obstacle in navigation 34 These works model multiple modality jointly but we instead focus on separate modeling since we do not find significant inter modality correlation during blockage Rather than direct fuse multiple sensor channels a few works utilize a secondary orthogonal sensory channel to as sist the main channel for better perception or sensing Girod and Estrin suggest using video evidence to solve the obsta cle problem in their acoustic ranging application 10 Qu et al add vision Pan tilt zoom camera and actuator pan tilt unit to help the LDV automatically select the best reflective surfaces point and focus the laser beam in order to remotely pickup voice signal 32 Stiefmeier et al
49. nd the produc ing wellhead Our experiment emulates oil blockages by controlling valves We are not able to inject actual blockages nor was it the time of year when they would form naturally Fig ure illustrates the topology of sensors pumpjack and valves Oval shape represents acoustic mote and squares do temperature ones T is located before the production circulation branch out and so upstream to both valves T and T are both on production line and straddle production valve downstream to T To emulate blockages we activate the production valve to close because it is not practical to cre ate a real blockage in the field When we close the production valve oil stops flowing in the pipe and hence we observe a total blockage in line with the valve In our experiment we always leave open either the production or circulation valve since closing both could cause high pressure at the wellhead that would damage the producing well or equipment Table 1 Experiment schedule and scheme product start pump valve purpose 11 0lam on Tj Jearn Ho open gt see 11 1lam off all learn u 12 01pm close T 2 non op 12 25pm on open T learn Ho 1 09pm T 2 in op close 1 54pm off all learn u 2 35pm open T learn Ho 3 05pm on close T 2 in op 3 48pm T learn Ho 4 30pm off open all learn u1 4 58pm on T learn Ho We conduct experiments on approximately half hour
50. next show the need to auto configure the parameters of our temperature algorithm and that with auto configuration deployment is robust to dif ferent wells and conditions We first show that baseline temperatures vary at different pipe locations and times and therefore we require different tuning parameters for different locations The left graph in Figure 8 shows the basic statistics over the entire 6 5 hour long temperature data while the right one focuses on the temperature under normal flow excluding no flow periods 340F 300 280 ca 1 i f L 260F mote ADC 240 K Y Figure 9 Box plot illustrate the fluctuation of temper ature when flow is normal at 7 during the experiment The x axis tick labels are the starting time of each event S e amp Y S 2 o 3 ROD Rays Ss oe oS m ue caused by either pump off or valve close We know that temperatures of pipes vary and they are affected by ambi ent temperature as well With only one day for field exper iments we cannot allow the pipe to completely cool In addition this statistics is not necessary because the over all and the normal flow statistics are enough to prove that temperatures vary We find that the temperature upstream to production circulation branch out 7 is significantly differ ent from the pair straddling the valve
51. nsing can not only re duce the cost of detection of cold oil blockages while avoid ing false alarms Automating sensing can provide much more rapid detection than current approaches Rapid feed back is important because a shorter gap between blockage reaches critical level and alarm is signaled can minimize dif ferent losses including environmental and equipment We detect blockage by sensing temperature and acoustic signals We infer flow interruption from pipe skin temperature but in addition to blockages many regular events change tempera ture including automatic pumpjack shut ins and diurnal en vironmental effects We avoid false positives by comparing multiple temperature readings and by using acoustic sensing to monitor pumpjack status We define our sensing problem and summarize our approach in Section Our experimental results focus on cold oil blockage but the principle of multi modal sensing to avoid false positives applies to many other sensing problems For example Girod and Estrin suggest using video evidence to correct prob lems from obstacles in acoustic ranging 10 In human mo tion detection Stiefmeier et al cross segment data stream between different sensors including inertial vibration and force sensitive sensors 36 We discuss more on generaliz ing our approach to other applications in Section 4 10 The first contribution of this paper is to identify the op portunity for multi modal sensing to reduce error
52. o detect pumpjack operation status Stoianov et al prove in PIPENET the feasibility of measuring vibration to detect small leak on water sewage pipe 37 However they do not present a completely integrated system no detection algo rithm is implemented Their field test only shows that their sensors deployed under urban sewage are capable of collect ing certain types of data and relay them back We instead focus on a complete multi modal system running oil line blockage detection algorithm online Jin and Eydgahi 17 and Sinha utilize acoustic wave propagation for pipe defects detection Jin and Eydgahi propose a general sensor network frame while focusing on specific signal processing analytical technique Sinha s work is mainly about trans ducer instrumentation and calibration for natural gas inspec tion Instead of pipe defects we focus on oil line blockage and real system development and deployment Contrary to forgoing low temperature fluid works we fo cuses on high temperature multi phase fluid oil gas water mix This fluid property change allows us to use tem perature for fluid presence detection 52 55 Zhu s work shows the feasibility of temperature monitoring for block age detection of pulverized coal injection system 55 Our fluid is mostly liquid while in his work fluid is actually fine powder His detection algorithm first measures temper ature by thermometers mounted on branch pipes differenti ates pipe s
53. or certainty of drop and the reference value k We set the threshold to 15 times normally observed temperature in this case 3 000 to be robust to transient temperature dips The reference value k is set as the mid point between quality level normal pipe temperature Ug and anomaly level flow stopped u1 41 lt Ho Since k is important to the accuracy and responsiveness of the algorithm we auto tune it instead of hard coding How ever due to different sources of noise we list in problem statement we do not think predefined fixed uo and u es timation can best reflect an appropriate k Hence it is nec essary to first auto tune uo and u for its dependency and we embed auto tuning in our algorithm to adjust the estimation of the two levels When pumpjack is operating determined by acoustic node introduced later in Section 2 4 we con stantly update the quality level uo by temperature observa tion When pumpjack shuts in we stop updating uo but start anomaly level u updating as temperature drops To general ize this we are using a second sensory channel to convert a false alarm hazard into a helper of parameter tuning By the time of the shut in is over and pumpjack resumes operating anew reference value k will be ready based on auto tuned to and u Another k tuning feature is that we do not update k at shut in because during pump off temperature detection be comes less important More importantly we intend to avoid
54. ory tests Section 4 9 The success of the both field and lab tests shows the generality of our approach on cold oil blockage and even a broader range of applications Earlier detection of partial blockages 50 or less would be helpful and is future work However our current temper ature method is not enough to detect the subtlety Possible future research could study how to implement sophisticated signal processing on sensor platform and how to leverage the differential temperature due to pressure change before and after blockage 2 2 Overview of approach To detect blockages we use two sensing methods acous tic sensing at the wellhead and temperature sensing at lo cations along the flow line Figure 2 Typical flow lines are much hotter than ambient temperature 100 C vs 0 30 C particularly in fields that use secondary production techniques such as steam injection We can therefore infer flow blockage by observing pipe skin temperature the pipe downstream of a blockage will converge to ambient temper ature In operation we expect to place multiple temperature sensors along the flow line near places where blockages are expected Unfortunately pumpjacks often stop production shut in periodically to allow downhole pressure to accumulate Pumpjack shut in causes drop in pipe temperature the same as a blockage so temperature sensing alone will result in false alarms We therefore add a second acoustic sensor to dete
55. ot ump bff k ump otf amp ptf Xap off 300 300 a 300 Q Q 200 a j 200 g lt 200 2 Loe a g Ea an CEA f 2 E 1007 y S 3 100 z se 100 3 h g g 8 5 g go g g 2 gg gilt L L L 1 L 1 a L L 1 L L gt 1 L 11 00AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM 11 80AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM 11 80AM 12 00PM 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM ye x i Ae x K gt x R Hae Ar 3000 Tf ny iri poog hy my in pae li A hi i haf 1 1 I bal i RE i rif Li V 4 if i Ly Ji vA i i ul u J V I L J if L L J L if T i h Ny T i i wane T i f a s truep i i s true i s truep 1 i i i a i 1 i 3 fi i i St 1 i Hen oy i LM 2b l i i false false false H TEREE i i i EE time a Upstream T b Downstream before production valve T c Downstream after production valve T2 Figure 12 Blockage detection results we take a three fold approach in this section We first for malize three simplified detection cases to theoretically prove our hypothesis Second we run simulation over these cases and finally we further support the simulation result by exper imental data The simplified detection cases consists of both acoustic and temperature modalities We employ the same acoustic model across the three cases that pumpjack status detection oscillates continuously at every observation time starting from to on off
56. our inexpensive sensors in the laboratory We next describe our field experimental setup and evaluation metrics test how temperature and acoustic sensors can infer blockages and equipment operation and finally show how their combina tion provides a robust system 4 1 Calibrating Individual Sensors Our premise is that low cost sensors are sufficient to de tect flow blockages We next compare inexpensive mote based temperature and acoustic sensors against high quality PC based sensors to confirm that inexpensive sensors are good enough 4 1 1 Temperature Sensor Measurement We show our acoustic mote is close enough to ground truth in previous section Next we compare temperature data by mote against USB data logger to verify if our low cost temperature collection solution performs well enough or not The major differences between the two systems lies in hardware and calibration The software processes are likely the same although only limited information about USB data logger disclosed by its manufacturer In our prior work 53 we find that in relatively low temperature range 0 200 C it is unnecessary to calibrate J type thermocouples before we deploy them in detection tasks Hence our mote reports raw ADC readings while USB data loggers are pre calibrated by manufacturer Our mote is almost equivalent to USB data logger because of the strong correlation between them pairewisely It is 0 91 on Ty 0 80 on ee and 0 83 on
57. pril june 2008 I Stoianov L Nachman S Madden and T Tokmouline PIPENET a wireless sensor network for pipeline monitor ing In Proceedings of the 6th International Conference on Information Processing in Sensor Networks IPSN 07 pages 264 273 Cambridge Massachusetts USA 2007 ACM S Tamura K Iwano and S Furui A robust multimodal speech recognition method using optical flow analysis In W Minker D Bhler L Dybkjr and N Ide editors Spoken Multimodal Human Computer Dialogue in Mobile Environ ments volume 28 of Text Speech and Language Technology pages 37 53 Springer Netherlands 2005 V Trifa L Girod T Collier D T Blumstein and C E Tay lor Automated wildlife monitoring using self configuring sensor networks deployed in natural habitats In Proceed ings of the 12th International Symposium on Artificial Life and Robotics AROB 07 Beppu Japan Jan 2007 Biometrika http bullseye xbow com 81 Support Support_pdf_ files MPR MIB_Series_Users_Manual pdf Crossbow mica 2 user s manual 2007 http bullseye xbow com 81 Support Support_pdf_ files MTS MDA_Series_Users_Manual pdf MTS MDA sensor board user s manual 2007 ttp datasheets maximintegrated com en ds AX4465 MAX4469 pdf MAX4465 MAX4469 data sheets 012 ttp www analog com static imported files ata_sheets AD1988A_1988B pdf AD1988A AD1988B alal o
58. r future work In addition we have a real deployment beyond theoretical models In short our major difference from prior work is our use of non invasive sensing Further we demonstrate a success ful wireless sensornet deployment in the field 5 1 2 Other Pipeline Monitoring Work Besides oil line blockage there are many other indus trial pipeline monitoring applications SCADA systems have long been used for pipeline monitoring Traditional SCADA systems use in situ sensors mostly single modal and cen tralized decision making 8 28 while our work instead fo cuses on detecting problems with heterogeneous intelligent low cost sensor in a network On the other hand prior sensornet research in pipeline monitoring usually assume low temperature single phase fluid 70957 Many researchers choose vibration sens ing effectively equivalent to our acoustic sensing for low temperature fluid monitoring NAWMS focuses on personal water usage 19 26 Our pumpjack status detection hard ware is similar to theirs in Mica series motes and MTS310 sensor boards However they use accelerometer while we use the microphone embedded on the same sensor board They infer flow rate by pipe vibration frequency and linear programming based algorithms We do not measure vibra tion in our flow inference because operating pumpjack gener ates wide band noise which overwhelms flow vibration sig nal Instead we use vibration as a secondary modality t
59. resence detection followed by acoustic pump jack status detection In addition to Accug we care about Accugn and Accuoff in acoustic sensing for future algorithm improvement 4 4 Accuracy of Flow Presence Detection We carry out full system deployment with algorithm on line in the field for both testing and data collection In order to incrementally test each component in the system we ran a manual version of temperature algorithm in parallel with the fully automated version on each mote The sole difference between the two versions lies in the process of parameter auto configuration we remotely re program the temperature motes to inform the manual version of the perfect pumpjack status while the automated obtains an imperfect update from their peer acoustic mote We use the manual version to eval uate flow presence detection while the fully automated is for blockage detection evaluation in Section The three plots in Figure 7 shows that our CUSUM based flow presence detection algorithm works perfectly during our field test We achieve 100 accuracy for the all ten events without any false positive or negative and we discuss more detailed observation below Since T is upstream to both pro duction and circulation valves oil flows as long as pumpjack is on regardless the status of the production valve Fig ure 7 a shows our algorithm remains silent while pump is on and temperature drops upon the first two pump off events effec
60. rnia United States Aug 1988 ACM Y Jin and A Eydgahi Monitoring of distributed pipeline sys tems by wireless sensor networks In Proceedings of The 2008 IAJC IJME International Conference Nashville TN USA Nov 2008 S Kim D Culler and J Demmel Structural health monitor ing using wireless sensor networks Berkeley Deeply Embed ded Network System Course Report 2004 Y Kim T Schmid Z M Charbiwala J Friedman and M B Srivastava NAWMS nonintrusive autonomous water moni toring system In Proceedings of the 6th ACM Conference on Embedded Network Sensor Systems SenSys 08 pages 309 322 Raleigh NC USA Nov 2008 ACM F Koushanfar S Slijepcevic M Potkonjak and A Sangiovanni Vincentelli Error tolerant multi modal sensor fusion In JEEE CAS Workshop on Wireless Commu nications and Networking Pasadena California USA Sept 2002 C Kreucher D Blatt A Hero and K Kastella Adaptive multi modality sensor scheduling for detection and tracking of smart targets Digital Signal Processing 16 5 546 567 2006 M Kushwaha S Oh I Amundson X Koutsoukos and A Ledeczi Target tracking in heterogeneous sensor networks using audio and video sensor fusion In Multisensor Fu sion and Integration for Intelligent Systems 2008 MFI 2008 IEEE International Conference on pages 14 19 Aug 2008 L Lazos R Poovendran and J A Ritcey Probabilistic de tection of mobile targets in hetero
61. s in oil viscosity as a result of cold weather sometimes compounded by buildup of sand in the pipe Blockage typically build up gradually over time Produc ing wells often operate intermittently with on off cycles of 5 15 minutes to allow downhole pressure to build up for suitable operation when the pump is not operational oil can transition from flowing slowly to blocked A blocked pipe can cause equipment damage and oil leaks since if well production continues with a blocked flow line pressure in the line will cause flow line rupture or pumpjack damage Recovery from equipment damage can easily amount to ten thousand dollars per event in addition to reducing produc tion Cold oil blockage is a significant problem in some oil fields Figure 1 shows eight consecutive years of produc tion data of an oil field where cold oil blockage is a con cern We normalize production values to remove long term decreasing trend in field production and show seasonal vari ation in production The first step of normalization is com puting monthly index by applying exponential decay fitting over the whole dataset The resulting fitting error is 1 3 low enough to show we have a good fit The fitting forecast is the monthly index the baseline of monthly production which is unaffected by the overall trend Next we normal ize the raw data by its ratio against its index for every month and the result is in the upper plot The lower plot summa 1
62. s large delay These approaches suggest the importance of the problem but sensor cost means they are deployed on only a few of the thousands of wells where there is concern Field engi neers confirm that in some cases the alternative is simply to preemptively stop production on certain well that have no monitoring Degree of Blockage Blockages build up over time and one would like to detect them before they happen or very quickly after they happen Currently pressure sensors are deployed on only a few wells Instead manual inspection is done to identify equipment damage that follows a blocking something that often occurs twelve hours after the fact Here we focus on rapid detection of full and near full blockages Detection of blockage allows well shut in and recovery before damage rapid detection after blockage may avoid equipment damage and will minimize leaks Our eval uation Section 4 7 shows we can detect blockage in 10 to 30 minutes much shorter than 12 hours by current manual inspection While not instantaneous this detection is poten tially able to save large production loss We emulate full blockages in field experiments Sec tion 4 We cannot test near full blockages in the field due to safety concerns but we do evaluate near full blockages in Figure 2 A diagram illustrates our problem statement Square symbols are temperature sensors the oval is an acoustic sensor S denote different pipe sections laborat
63. ther sensing modality for better cost efficiency For ex ample from our survey we find some potential alternatives are infrared imaging and vibration Finally detecting partial blockages is future work 5 Related Work Our work builds on prior research results in flowline mon itoring change point detection algorithms and multi modal sensing 5 1 Pipeline Monitoring Systems Our work first builds on pipeline monitoring related works especially in oil line blockage detection 5 1 1 Oil Line Blockage Detection Applications Oil line blockage is not a new problem to the oil industry We study how to use multi modality with low cost sensors to detect line blockages but prior work has looked at alternative methods Liu et al shows it is practical to locate blockage in a long oil line by measuring the travel time of pressure decompres sion wave bouncing back from the blockage point 24 They test their method by emulated blockage by an intermittently operating oil line and we did similar blockage emulation for testing However their sensing pressure modality is invasive which is different from our non invasive sensing Liu and Scott show a theoretical work to localize blockage in subsea flowlines by comparing the inlet and outlet line pressure and other factors They use invasive pressure sensing while we are non invasive Our work suggests that we can locate blockage by segmenting pipe with pairwise temperature sensors which is ou
64. tion is not effec tive is that our acoustic algorithm runs with fixed threshold For example the noise floor rises in the third pump off pe riod 4 30pm 4 58pm triggering false positive under now too low threshold We are currently working on making our algorithm adaptive to cope with this situation In addition to an adaptive algorithm this result suggests that filtering noise during pump off may too improve pump off accuracy or overall To verify if noise filtering helps we next apply our algorithm to a cleaner dataset by PC We find the easiest way to improve the accuracy of pump off detection is to upgrade the hardware for acoustic mea surement because the same algorithm works perfectly on PC acoustic dataset Figure I shows the same experiment as forgoing but collected by PC microphones We take six min utes of the data covering the first pump on off transition to configure the threshold with the same auto configuration al gorithm The end detection result is encouraging with 100 300 acoustic N So mote ADC pasoo ATLA EDSO ATEA 1 1 L l 2 00PM 3 00PM 4 00PM 5 00PM ot L 1 i 11 00AM 12 00PM 1 00PM EV r z x 7 i l l L x time Figure 10 Acoustic pumpjack status detection result by mote Pump status is denoted at the top of both plots Dashed dotted vertical lines indicate the production valve is open closed The tags v
65. tively triggers our algorithm In addition our algorithm successfully detects temperature drop caused by in line valve close up showing in Figure 7 b and 7 c where C4 builds up at all valve closed events We expect to see a trigger in the third pump off event 4 30pm 4 58pm marked but surprisingly Cy does not build up high enough in the algorithm different from the prior two pump off events We still count that a true posi tive rather than a false negative because it is a result of our compressed experimental schedule we ran out of the time at the end of our experiment and hence we cut off the third pump off event prematurely The trend shows we require 5 400 a 300 pi T _ Q T l T gl Aa o 200 i aE 2 T 100 oly all data normal flow only 0 Ty To Ti T T TG Figure 8 Two box plots illustrate the difference in tem perature between three location The blue boxes cover both the upper and lower quartile with a red median mark in the center The whiskers extend to 1 57 in terquartile range excluding outliers red marks additional minutes to trigger and since operational condi tions do not have a 30 minute time limit we count this event as correct Two evidences are that the temperature still main tains a steep drop trend and Cy does start to build up at T and T After comparing t
66. ubing clang However the major difference between pump on and off signal lies in the average amplitude pump off is much quieter relative to ambient noise while pumpjack flow is relatively loud The result shows with proper parameters our algorithm is capa ble of distinguishing average amplitude difference between pump status In this test we use a longer cycle 18s and lower pump on threshold 60 percentile of amplitude among pump on training samples comparing to our field test On the other hand the success of our acoustic detection on sig nal with different properties than pumpjack generalize our approach to a broader range of applications Combining the perfect temperature and acoustic detec tions the end near full blockage detection accuracy is 100 over the total of eleven events The lower plot of Figure shows that our algorithm triggers on all five blockages and the two pump off periods are correctly suppressed In all our lab test shows our multi modal sensing works with parameter changes on near full blockage detection in a hot water network We believe this result generalizes to 0 03 1 1 acoustic pump off pump j 0 v aa lt I lt lt lt I lt P Zook je pE E ESE oe z ait S js ig S s is S 8 is l 4 gigs NS conn RS RQ ie a 0 01455 i R ys A HIRIS o y inal FF gut 1 L fi 1 f L im 1 00PM 2 00PM 3 00PM 4 00PM 5 00PM 6 00PM 7 00PM a on
67. ur initial field trials show if exposed under the sun directly temperature sensors with the amplifiers sometimes return random readings but sensors without the amplifiers work correctly Hence in the latest test Section 4 2 we covered the sensor motes in shade but we are currently ex amining our design and seeking a more robust solution 3 2 Hierarchical Sampling and Aggregation in Acoustic Mote To obtain the sound pressure level of pipe our acoustic sensor samples 2000 times a second This sampling rate is high for a mote posing two challenges First although the sensor generate and transmit one packet per second we cannot collectively stack 2 000 one second long samples in buffer due to the limited Mica 2 RAM size 4kB for both program and data The other challenge is that because the sensor samples at such short interval as 500 us hardware in terrupts from other components radio flash logger etc are likely to cause large variation in sampling rate 14 18 For accurate sampling we shut down all external components which might occupy the CPU for too long to hold up the timer Hence we design our software able to schedule and interleave processing transmitting and flash logging among continual sampling We do local flash logging because in operation it could serve as backup in case of temporary net work outages although in a fully integrated system data is always streamed back to a central server through field net
68. urns values ranging from 0 to 1023 mapped to 0 to 3 V As a result it does not return negative voltage To avoid losing the nega tive half of the waveform MTS310CA is designed to elevate the center of the output acoustic waveform from 0 V to ap proximately 1 5 V which corresponds to 512 in ADC value We test this feature with our equipment and find that the new ADC waveform centers around 501 slightly off by the theo retical value of 512 We thus use our experimental result to offset mote ADC readings removing DC bias Second we decouple microphone unit from the board for better mounting The flat Mica sensor board does not well match the curved pipe surface Therefore we desolder the microphone off and extend it out via wire which enables us to simply tape it down to pipe in deployment for best contact and windscreen Finally we use TinyOS to maximize the microphone ana log gain We use the OS service to tune an resistor in the amplification stage to its largest value which is an on board digitally controlled variable resistor 41 4 Evaluation We next describe the experiments we carried out to demonstrate we can detect flow blockage and that multi modal sensing can avoid false positives We first evaluate a Acoustic mote with microphone extended b Mote mic on a pipe c Temperature mote d Thermocouple on a packed in a box pipe Figure 3 Our temperature and acoustic sensor hardware and deployment
69. w presence detection is the first part of our multi modal cold oil blockage detection In this sec tion we talk about how to detect flow presence by tempera ture and how to automatically tune parameters According to our problem statement and hypothesis above we need to measure pipe skin temperature to detect the presence of flow or in another words suggested block age Since the temperature usually drops gradually about 20 C in an hour we need an algorithm to process stream ing temperature trace and identify its trend of approximating ambient For the above reason our algorithm employs one sided CUSUM or cumulative sum control chart 31 originally a statistical technology developed for process quality con trol The algorithm starts at low pass filtering raw tempera ture observation by EWMA to filter transient noise Next it compares every observation to a reference value to calcu late the deviation from it Meanwhile it maintains a running Statistics the cumulative sum of all the deviation in history as basic CUSUM does In this paper we call this cumulative sum of deviation certainty of drop C4 When observation is lower than k C4 becomes larger and larger before it exceeds a threshold which suggests a blockage because the temper ature is too low for too long We use one sided CUSUM resetting Cy when it is less than zero to respond quickly to temperature drops We must set two algorithm parameters the threshold f
70. we and T showing by the different quartiles blue boxes in either plot Hence auto configuration is critical because one parameter setting works on one location does not necessarily work on another For example a reference value of 229 ADC value gives 100 accuracy on T but would trigger three false positives on sensor downstream to production valve T2 Additional motivation for auto configuration is that tem perature changes constantly at the same location Therefore hand tuning parameters on each sensor to cope with the lo cation disparity is still insufficient Figure 9 breaks down the T temperature trace and compares between eight events only when flow is normal The temperature measurement fluctu ation is significant mostly caused by a noise combination of diurnal amplitude downhole change and back pressure Section 2 3 The first and last three boxes has no over lap with the other four which suggests that maintaining a fixed threshold during the whole time is likely to cause mis detections Our further study confirm with this observation and hence an adaptive temperature algorithm is necessary To address the above problems our flow presence de tection auto configures the most important parameter the CUSUM reference value based on the training tempera ture under normal flows and pump shut ins details in Sec tion 2 3 Our 100 accuracy shows it is effective Sec tion 4 4 More importantly we find it is n
71. work We design a hierarchical sampling and aggregation scheme to overcome the two challenges above Overall we pause the high frequency sampling and schedule other oper ations before next sampling cycle The pause causes gaps in sampling and in the worst case we may mis detect interest ing phenomenon To minimize this sampling gap and coor dinate data management we make following design choices At a high level our sensor samples and computes the SPL within a one second long window long window before log ging it to flash and transmitting it out At an intermediate level we divide each long window into ten 0 1 second long short windows In each short window sensor samples for 0 06 s at 2 kHz rate and uses the remaining 0 04 s to do SPL aggregation The final 10 short window does further aggre gation by choosing the maximum SPL value among the past ten to represent the entire long window before flash logging and radio transmission Our lab testing shows 60 40 duty cycle is optimal because a slightly more aggressive setting i e short than 0 04s gap causes significantly more packet loss Besides the 0 04 s gap does not cause mis detection on the 0 2 second long signature rod tube clanging noise 3 3 Maximizing Acoustic Gain To maximize the acoustic signal gain we take three steps on software and hardware customization First we optimally adjust digital current bias through cal ibration The 10 bit ADC channel of Mica 2 ret

Reducing False Alarms with Multi

Contents

Download Pdf Manuals

Related Search

Related Contents