using an accelerometer to increase remote control safety
Contents
1. NDO Jumping down small heights 80 cm and walking in stairs These shall not be misinterpreted as falling or stumbling or the remote being dropped TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 24 ND1 The operator is standing still or sitting with the remote on his lap This shall not be confused with the remote being left unused 3 2 Technical requirements This section examines what is technically required in order to meet the above requirements This has been done by looking at the patterns of acceleration of an event The patterns were derived by sampling the output data while executing the use cases of the functional requirements Section 4 2 discusses the algorithm that implements the required functions Note that a belt was used for the prototype which didn t prevent it from bouncing up and down The result is a bit of bouncing noise that coupled with the output data Every case was tested both with and without holding the remote still and no significant difference was found in the characteristics Only the overall acceleration magnitude with bouncing is higher in some cases Also the prototype is lighter so that it is accelerated even faster than the real remote DOT Operator falls This case has been solved in AN 1023 and is discussed in 2 4 1 D1T The remote was dropped from 30 cm height into a cardboard box padded out with foam Graph 2 shows the significant elements that are already known from DOT T2. 5 Conclusion The required knowledge was gathered and accelerometers in particular the ADXL345 have proven suitable for the application and the detection and differentiation has been implemented successfully Although not every case was implemented the goal was achieved because it was examined how this can be done see 5 1 One drawback is that a complete separation of normal and abnormal situations was not achieved Due to the required sensitivity it was not feasible to avoid some false detection in the first place However when using the application in the final product the control can be re enabled right after a detection has proven to be false alarm 5 6 Future improvements Due to a lack of time the present solution does not provide the best possible usability since it was the goal to detect all situations Before it can be used in the product the thresholds and times of the algorithm have to be adjusted to increase usability and reliability Furthermore since the ADXL345 offers many specific features that haven t been the focus of this project they could be used in the future For instance the part s FIFO can be used in order to reduce host processor burden In order to save power low power mode sleep mode and link mode can be used In link mode activity is only detected if inactivity has been detected last and vice versa 2 16 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 45 Also it might be g
3. Output response vs Orientation to Gravity 2 Data format and retrieval The ADXL345 continuously updates its data registers with a user specified output data rate up to 3200 Hz The format of acceleration data is twos complement and for each axis it is separated in two 8 bit registers one holding the lower bits and the other one holding the higher bits and the sign bits The amount of sign bits depends on the selected output resolution which can be up to 13 bit For post processing the register values need to be assembled after reading It is recommended to read the data registers with a multiple byte read so that their content isn t updated between reading operations Figure 3 shows the data format in full resolution mode which has been used during this project Configured with resolution of 13 bit and a range of 16 g the z axis should read a value of 256 when in 1 g field see Theory of operation ors o14 013 p12 prs 010 o os o7 oe os bs os 02 os oo sion sien sien sien 011 10 v9 ve 07 ve os os os o2 o1 Do DATAX1 DATAXO DATAY1 DATAYO DATAZ1 DATAZO Figure 3 Format of data registers in 13 bit mode 4 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 13 Functions While acceleration can be read directly from the data registers it is also possible to use 8 different interrupts see paragraph Interrupts that provide advanced state information as well as a self test
4. a small acceleration threshold is used to indicate activity Another application is with a bigger threshold to recognize impacts This TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 14 interrupt can be used in AC or DC coupled mode By default the DC coupled mode is used i e the measured acceleration is compared to the specified threshold This mode is used when the magnitude of acceleration is relevant e g in case of an impact In AC coupled mode again interrupts are generated based on the magnitude of the change in acceleration 8 For instance this can be useful if a watch is used to detect constant wrist movement Inactivity The Inactivity interrupt works similar to the Activity interrupt It occurs when acceleration is less than the specified threshold for longer than specified by the time threshold As with the Activity interrupt the participating axes can be configured This makes it more adjustable than the Free Fall interrupt Additionally AC or DC coupled mode can be chosen which works the same way as for Activity In AC coupled mode a reference value is taken at the beginning of inactivity detection and compared to the current acceleration An interrupt is generated if the difference is less than the inactivity threshold for the time specified in the time parameter The result is that the magnitude of the change in acceleration matters and not the magnitude of acceleration In DC coupled mode the
5. above state machine and how it meets the technical requirements 3 2 is described in detail DOI Operator falling down An operator fall is detected in the same way as in the original algorithm which is described in 2 4 1 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 35 D11 Remote being dropped If the use case operator jumping down is compared to operator falling the difference is that the operator turns over during a fall resulting in a status that is different from the initial status However this criterion doesn t differ between operator jumping down and remote being dropped compare 3 2 D1T and NDOT so that the following sequence of detection statuses is representative for both Weightlessness status 0xF1 Impact with ground OxF2 Stability OxF3 Start over since the position equals the initial position OxFO ge a CI A restart 4 is appropriate after jumping down but not after the remote was dropped In order to tell the events apart it is necessary to look at what happens afterwards The following is an idea how these cases can be differentiated It hasn t been implemented because it s an advanced feature and the goal was not a product ready application Lying on the ground the remote will most likely remain completely motionless which leads to detection as left unused sooner or later depending on the specified inactivity time However since the remote is not in ha
6. detect behavior Task delay has passed since the last time that we were here So increment timer variable What is the DetectionStatus React accordingly if intl adx1345 read_reg XL345_ INT SOURCE amp ADXL345Registers XL345 INT SOURCE 1 What type of interrupt has occurred What is the DetectionStatus React accordingly Figure 11 Structure of the function derived from the original interrupt handler TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 32 abnormal det h define DELAY XL 20 define POSITION WINDOW 500 DELAY_XL DELAY XL ms LSB define STRIKE _ WINDOW 200 DELAY_ XL DELAY XL ms LSB define STABLE WINDOV 3500 DELAY XL DELAY XL ms LSB define FREE FALL OVERTIME 300 DELAY XL DELAY XL ms LSB define FREE FALL INTERVAL 100 DELAY_XL DELAY XL ms LSB Figure 12 Correlation between task delay and time windows jpportTASK FUNCTION vAbnormalTask pvParameters detect behavior vTaskDelayUntil amp xLastWakeTime DELAY XL Figure 13 DELAY_XL is used both as the task delay and to calculate correct values for the time intervals of the algorithm The state machine Figure 14 shows the states of the original algorithm which were used for the solution State transitions with red letters ind
7. feature see paragraph Self test a FIFO see paragraph FIFO and the possibility to calibrate the offsets of the axes of acceleration see paragraph Offset calibration Interrupts The ADXL345 provides 8 interrupts that can be used in different ways to suit specific user needs Each of them participates in the Interrupt Source register only if the respective bit is enabled in the Interrupt Enable register For more details on interrupt usage and parameters see 2 and 4 3 Parameters Subsequently the main characteristics of these interrupts are described Free Fall During a free fall an object experiences weightlessness In accelerometers this state is represented by an acceleration of 0 g on all axes Despite its name the free fall interrupt can be used to indicate falls in general not necessarily free falls 8 There are two parameters that must be set in order to use the free fall interrupt an acceleration threshold and a time threshold The part will generate an interrupt when the current acceleration has been below the threshold for more time than specified in the time parameter Activity It is possible to enable or disable each one of the x y and z axis to participate in the detection of activity When the magnitude of acceleration on any participating axis exceeds a specified threshold an Activity interrupt occurs The activity interrupt can for instance be used in combination with the inactivity interrupt whereat
8. in detail how it was done using a state machine with detection statuses 4 2 1 Overview The result is based on the algorithm described in Application Note 1023 which has the shape of a state machine see 2 4 1 and 3 The technical requirements see 3 2 are realized through the accelerometer s interrupts see section Interrupts whereat the Activity Inactivity and Free Fall interrupts were used While the original application only raises an alert after a valid sequence of events was detected see Figure 14 in the current project a reaction must occur as soon as possible The reason for this is that no more commands shall be sent to the machine in case of an abnormal situation Although the goal of this project was only the detection of abnormal behavior but not the implementation of appropriate reactions the places for a later reaction in the TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 29 product ready application are indicated by the functions enableCtl and disableCtl They set LEDs and buzzers on the remote in order to indicate the reason for the reaction in this case the reaction is enabling or disabling of the control The only requirement for an implementation of appropriate reactions is that each situation has a unique characteristic on the base of which it can be distinguished from others 4 2 2 explains how this is done However for the final product further decisions about how to react will be nec
9. invalid the control is disabled If it is valid again the control is re enabled Like in the initial state weightlessness shock and motionlessness are also detected while the position is invalid If the part didn t detect a fall or strike while the position was invalid and it ended up lying on the ground in a valid position the algorithm would start over with the initial state see Figure 15 In the product ready application this would re enable the control which could be dangerous because the user might not have the control e g because he lost consciousness due to an accident or because he has dropped the remote Also it is no valid option to rely on the detection of motionlessness left unused which turns off the control after a while because there would still be a period within which the device could transmit unwanted input to the machine This is to be avoided in a safety critical application The parameters are the interval when the position is checked and a threshold that controls how much the position can differ in order to be still valid see 2 4 2 D5I Losing balance This case is covered by DOI because it may lead to a fall which will be detected in time Again if the operator regains balance and doesn t fall there is no need for a reaction TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 38 NDOI Operator jumping down This case is covered by the sequence of detection statuses mentioned in D1
10. magnitude of acceleration is compared directly to the specified threshold see Activity Other interrupts The ADXL345 also provides additional interrupts that have not been used in this project One of them occurs when a new sample of output data is available Data Ready interrupt Another one indicates that acceleration greater than a threshold occurred for less time than a time parameter Single Tap interrupt The Double Tap interrupt includes two single taps separated by a user specified time Tap detection can be useful for applications where it is desired to sense tilts for instance motion enabled video games or cell phones TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 15 The Watermark and Overrun interrupts are primarily useful in combination with FIFO and are not discussed here since FIFO is not used in this project Self test The self test feature provides the possibility to test the accelerometer by moving the mechanical parts in the same way as acceleration does Self test should be used before a new accelerometer is used for the first time in order to assure that it is functioning correctly This test is conducted by measuring two sets of acceleration data the first measurement is done with the self test feature disabled and afterwards compared to the data measured with self test enabled If the difference is within a threshold the test is considered successful The thresholds depend on supply voltage temper
11. 1 It can be considered as successfully not detected since the algorithm starts over right after the jump ND1I Operator standing still or sitting with remote As pointed out in chapter 3 it is not straightforward to distinguish whether the remote is left unused D21 or the operator is sitting or standing still Due to a lack of time this case was not implemented but it can be done like this 1 A high value for inactivity time can be used This increases the probability of detecting minor motions while the operator is sitting or standing 2 However if he remains completely motionless the detection algorithm could be interfaced with remote input e g joysticks buttons to determine that the control is still being used 3 Another option is to set the acceleration threshold of the activity interrupt very low after long time motionlessness has been detected by the inactivity interrupt If the activity interrupt occurs within a specified amount of time use a timer there is still movement and the remote is not left unused If however no activity is detected within the specified time interval the remote is identified as left unused A suitable value for the activity threshold can be determined by trial and error 4 3 Parameters This section deals with the parameters used to adjust the algorithm The main parameters for adjusting the algorithm are the parameters of the three interrupts that it uses Activity Inactivity
12. 1 Hardware 2 1 1 The remote control 2 1 2 ADXL345 accelerometer 2 2 Software 2 2 1 FreeRTOS 2 2 2 Compiler Sourcery G Lite ARM EABI 2 2 3 Serial communication PUTTY 2 3 12C serial bus protocol 2 4 Existing work 2 4 1 Application Note 1023 Fall detection 2 4 2 Competitor solution 3 REQUIREMENTS 3 1 Functional requirements 3 2 Technical requirements 4 REALIZATION 4 1 Console program 4 2 The algorithm 4 2 1 Overview 4 2 2 Design 4 3 Parameters 5 VALIDATION AND DISCUSSION 5 1 Test results 5 2 Response time analysis 5 3 Personal learning experience 5 4 Employer s feedback 5 5 Conclusion 5 6 Future improvements REFERENCES TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern o O ON Oo 17 17 18 18 18 19 19 23 23 23 24 27 28 28 28 30 38 40 40 42 43 44 44 44 45 PICTURES Picture 1 Technion TRC remote control left and the used prototype right 10 GRAPHS Graph 1 Acceleration change curves during the process of falling 3 20 Graph 2 Acceleration change curves of remote control being dropped 25 Graph 3 Jumping down from chair sess ciateetcehi cies cabeckencece aioe wetdenndecbitnhenqechde eacseraes 26 Graph 4 Standing still with remote c eee ee cece cece e ee eeee eee a eens sete eetennaaeaeeeeeeeeeenaa 27 FIGURES Figure 1 Axes of Acceleration ADXL345 data sheet 0 0 eeccceeeeeeeeesteeeeeeeeeeeeeeeteaaees 11 Figure 2 Output res
13. After the calculated offsets have been written to the offset registers the part automatically compensates for any offset Still the measured acceleration should be verified with the expected values which depend on the part s sensitivity 2 Initialization In order to function correctly the chip requires a minimal sequence of initialization which is shown in Figure 4 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 17 START DESCRIPTION 16g 13 BIT MODE START MEASUREMENT ENABLE DATA_READY INTERRUPT AME oara DATA_FORMAT POWER_CTL INT_ENABLE Figure 4 Minimum initialization sequence 4 2 2 Software 2 2 1 FreeRTOS FreeRTOS is a free professional RTOS that supports a variety of microcontrollers This project s functionality was developed and tested in a FreeRTOS task which will be used in the final product Subsequently the features that have been most important for this project are described In FreeRTOS time is expressed in ticks The functions vTaskDelay and vTaskDelayUntil are the two options to block a task so that the CPU will be available to the task with the highest priority among the tasks that are ready to run vTlaskDelay blocks the task for a specified amount of time while vTaskDelayUntil blocks the task until an absolute point in time is reached For time measurement e g in terms of response time analysis see 5 2 the xGetTickCount function was
14. Bachelor s thesis Information Technology Embedded Software 2012 Lukas Kern USING AN ACCELEROMETER TO INCREASE REMOTE CONTROL SAFETY TURUN AMMATTIKORKEAKOULU TURKU UNIVERSITY OF APPLIED SCIENCES BACHELOR S THESIS ABSTRACT TURKU UNIVERSITY OF APPLIED SCIENCES Information Technology Embedded Software Completion of the thesis 6 2012 Total number of pages 39 Instructor Lic Tech Jari Pekka Paalassalo Author Lukas Kern TURUN AMMATTIKORKEAKOULU THESIS This thesis deals with the usage of an accelerometer in order to determine the status of a remote which is used to control vehicles such as for instance excavators While using the remote the user may drop it or stumble which may lead to unpredictable input that shall not be transmitted to the machine The situations that produce such input were analyzed in order to be capable of detecting them and reacting accordingly Although not every possible scenario was discussed or implemented it has been shown that accelerometers are suitable for increasing the safety of remote controls KEYWORDS ADXL345 accelerometer remote remote control radio controller wireless movement safety emergency machine vehicle excavator crane machine safety FreeRTOS UART I C signal processing sampling FIFO TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern CONTENT LIST OF ABBREVIATIONS OR SYMBOLS ACKNOWLEDGEMENTS 1 INTRODUCTION 2 BACKGROUND INFORMATION 2
15. FFERENT FROM INITIAL STATUS GENERATE FALL ALERT ACTIVITY INTERRUPT ASSERTED NO NO INACTIVITY INTERRUPT ASSERTED YES GENERATE CRITICAL GENERATE FREE FALL ALERT CRITICAL ALERT Figure 8 Flow chart of the original algorithm 3 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 22 void IRQ Handler irq IRQ interrupt if timer interrupt Increment timer variable Threshold exceeded What is the DetectionStatus React accordingly if IRQSTA amp SPM4 IO BIT SPM4 IO BIT External interrupt from ADXL345 INTO What type of interrupt has occurred What is the DetectionStatus React accordingly Figure 9 Code structure of the original algorithm The application receives an interrupt either from a timer or externally from the interrupt pin of the ADXL345 as can be seen above The interval for the timer interrupt is 20 ms and if it occurs and for instance weightlessness has been detected a timer variable is incremented to wait for an activity interrupt that signals impact with the ground If this interrupt is not detected in time this indicates an invalid status and the algorithm starts over This process is displayed in Figure 10 TimerWaitForStablett if TimerWaitForStable gt STABLE WINDOW Time out restart Figure 10 Role of timer variables in detect_behavior Inclination position check After weightlessnes
16. and Free Fall see paragraph TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 39 Interrupts The next section motivates the values and for the Activity and Inactivity interrupt the mode of operation that were used Activity As mentioned in Interrupts the Activity interrupt can be used in AC or DC coupled mode In the initial state as well as after a free fall has been detected the part is looking for a strike and is therefore used in DC coupled mode AC coupled mode is used to detect movements while waiting for stability after a strike as well as after weightlessness followed by a strike Inactivity This interrupt is only used in AC coupled mode for stability detection after a fall or strike as well as for detecting whether the remote is left unused The used value for the acceleration threshold is the same in every case while the time parameter is shorter for the detection of stability OxF3 than for motionlessness OxFA Free fall The closer the acceleration threshold is to 1 g which is the acceleration sum during motionlessness the easier it is to detect a fall On the other hand this may cause other movements to mistakenly trigger the interrupt A value close to 0 g again will make it easier to distinguish a fall from other movements In this case however a fall that doesn t involve complete weightlessness will not be detected A suitable acceleration threshold highly depends on the specific appli
17. and retrieval and the C transmission protocol was used to communicate with the part see 2 3 12C Theory of operation In order to measure acceleration an accelerometer applies Newton s second law of motion F md It measures the force that acts on an object with a TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 11 known mass called a proof mass The most common way to measure this force is by measuring the dislocation of the proof mass 12 Figure 1 displays the axes of acceleration sensitivity whereat the corresponding output voltage increases when accelerated along the sensitive axis For instance consider the case where the part is lying flat on an even surface Depending on which side of the part is up and which is down the z axis is in the 1 g field of gravity and reads 1 g and the x and y axis are in the 0 g field of gravity and hence read 0 g This can be seen in Figure 2 It also shows how to position the part so that the x and y axis are in the 1 g field If the accelerometer remains in one position and isn t moved each axis experiences a magnitude of acceleration between 0 and 1 g depending on its alignment with the earth s field of gravity Az Figure 1 Axes of Acceleration 2 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 12 GRAVITY fp it dOl e Xout 1g Your 9g Zour 09 Xour 0g Xour 0g Your 0g Your 9g Zout 19 Zout 19 Figure 2
18. ature output resolution and output range FIFO A FIFO short for First In First Out is a data structure that works like a queue By default the FIFO is not operational If the FIFO is in use new data is placed in the FIFO but it s still accessed via the data registers A possible FIFO application is to reduce host processor burden by setting the part to sleep mode until the FIFO has filled to a user specified extent and then waking up and reading the cumulated samples in a burst The FIFO holds up to 32 samples of data for each axis and can be used in three different modes These modes are compared in Table 1 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 16 FIFO mode The FIFO collects data until it is full Stream Continues to collect data when full keeping the latest samples and mode discarding the oldest Trigger First holds only latest samples stream mode After a trigger event it mode operates in FIFO mode Trigger mode can be used if axis data right before an event interrupt is of interest Table 1 FIFO modes of operation Offset calibration Before the ADXL345 will return correct values of acceleration offsets must be calibrated for its axes This is due to mechanical stresses that can occur when the part is physically installed in i e soldered to a system Offset calibration can be done by measuring acceleration while the part is lying flat on a surface see Theory of operation
19. cation In the original algorithm a rather big value of 0 75 g has been used so that any falls can be detected 3 In this project though the user moves a lot and the remote bounces up and down occasionally In this spirit a smaller threshold would prevent normal movements from being confused with safety critical movements On the other hand the detection of falls would be less reliable with a smaller threshold i e some would not be detected Therefore it must be TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 40 tolerated that the control is temporarily disabled when bouncing up and down since safety has the highest priority 5 VALIDATION AND DISCUSSION Section 5 1 deals with the functional tests and their results and section 5 2 lists the response times for the solution Section 5 3 reviews the project and experiences from my side and 5 4 from the employer s point of view Section 5 5 presents the conclusion and 5 6 hints at possible improvements for the future 5 1 Test results The following tests show that the functional requirements have been met Note that before the application can be used in the remote product more extensive tests and adjustments are necessary DOV Operator falling down A typical sequence of detection statuses looks like the following Weightlessness Timeout Invalid position Free fall Impact Stability Invalid position N Oo OD At first the fall times out whi
20. ccur i e the respective timer times out TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 30 4 2 2 Design The detection of abnormal behavior is implemented as a function detect_behavior that is called by a FreeRTOS task The microprocessor doesn t receive interrupts directly from the interrupt pin of the ADXL Instead polling was used to determine whether this pin was high or low The reason is that if an interrupt handler was used the part could generate an interrupt flood which would prevent other functions from running Polling grants better control over the CPU time that is consumed by the accelerometer Figure 11 shows the derived function which uses polling In the original algorithm a timer interrupt occurs every 20 ms and the values of time windows are calculated based on this interval Since no interrupt handler is used in the current application the same time interval was used both as the task delay and in the algorithm parameters to calculate the time windows The interconnection between algorithm logic and task delay might seem like a disadvantage However since the task delay is constant it s more suitable than using for instance time stamps in detect_behavior Figure 12 shows how the time windows for the algorithm depend on the delay and Figure 13 shows how the detection function is used in the context of a FreeRTOS task TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 31 void
21. d adapted in this project see 4 2 The algorithm The authors determined four characteristic elements of a fall A valid fall situation is only diagnosed if all of these have occurred Graph 1 shows the belonging output data 1024 1 WEIGHTLESSNESS Se VECTOR SUM 2 IMPACT X AXIS 768 3 MOTIONLESS om Y AXIS 4 INITIAL STATUS Z AXIS 512 D 1 oS wo y m 256 I 3 m I a i gt l 0 l i 256 512 1 51 101 151 201 SAMPLES 50 s Graph 1 Acceleration change curves during the process of falling 3 Here weightlessness means that the acceleration on all axes goes towards 0 g graph 1 phase 1 It is followed by an impact with the ground phase 2 and then by motionlessness after the fall phase 3 The fall involves that the test person turns over so the acceleration before and after the fall is different compare phase 3 and 4 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 21 The algorithm uses 3 interrupts Activity Inactivity and Free Fall Further it uses detection statuses to indicate for instance weightlessness motionlessness or impact with the ground Hence it can be compared to a state machine Figure 8 shows the flow chart and Figure 9 the code structure of the original algorithm INITIALIZATION N FREE_FALL INTERRUPT ASSERTED YES CONTINUOUS FREE_FALL DETECTED NO NO ACTIVITY INTERRUPT ASSERTED YES NO INACTIVITY INTERRUPT ASSERTED STABLE STATUS IS DI
22. data rate of 100 Hz 16 g range and 13 bit output resolution see Data format and retrieval and appropriate values for the offset registers were determined and are written to the registers every time the device is powered on see Offset calibration They need to be determined only once after an ADXL has been installed on a system A console program was written to facilitate communication with the device and is described in 4 1 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 28 A fall detection algorithm from an application note which is described in 2 4 1 turned out to provide a good basis for the detection of abnormal behavior It shows how the ADXL345 interrupts can be used to obtain the desired status information about the part In most cases their usage makes it needless for the application program to process acceleration data directly The adaptation of the algorithm is described in 4 2 and the relevant parameters to adjust it in the future are explained in 4 3 4 1 Console program To facilitate the process of development a console program was written which can be used in order to read and write the registers of the accelerometer Furthermore it can sample acceleration data and print it to the console As a help to test the algorithm the console program can display a log of the latest detection statuses which are discussed in 4 2 2 4 2 The algorithm 4 2 1 gives an introduction to the algorithm and 4 2 2 explains
23. ed implementation to be not detected implementation to be detected validation to be not detected validation Millisecond s Nanosecond s Microsecond s TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern ACKNOWLEDGEMENTS First of all I d like to thank Timo J skel inen for introducing me to this project and providing me with the opportunity to make valuable technical and interpersonal experiences couldn t have done this project without the patience and advice of the colleagues from Technion Next ld like to thank my supervising tutor and teacher Jari Pekka Paalassalo for his thorough guidance in finding a thesis topic and during the work process In the same way I d like to thank my employer Antti Pasanen who took the time and effort to give me guidance and valuable feedback despite a lot of other tasks on his list Many people that met during my student exchange inspired me and made my stay in Finland a great experience want to thank Roni for having been stubbornly motivating Till for his support and many inspiring conversations and Davide for reminding me how important it is to never give up Also I m grateful for the love from my mother my father and Gela and my brother Niklas who helped me to continue when felt demoralized TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 1 INTRODUCTION When the goal of an application is to provide safety this requires never ending analys
24. eeeeenaeeeeeeeeeeeeeeesaaees 34 TABLES Table 1 FIFO modes of operation cccceeeeeeeseesnceeeeeeeeeeeeeenaeaeeeeeeeeeseseeaaaeeeeeees 16 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern LIST OF ABBREVIATIONS OR SYMBOLS Remote Control CPU FIFO ISR EABI g RTOS SDA SCL CMOS GPIO ODR AC DC l2C GNU Project Toolchain R W LED Hz D number short for remote control refers to the control over the controlled machine not to be confused with remote Central processing unit First In First Out data structure equivalent to a queue Interrupt service routine interrupt handler Embedded application binary interface Gravity of Earth 9 81 m s Real Time Operating System Serial data signal serial clock Complementary metal oxide semiconductor General Purpose Input Output Output Data Rate of the ADXL345 Alternating Current Direct Current Inter Integrated Circuit serial bus A free software mass collaboration project A set of programming tools such as compiler and linker Read Write light emitting diode Hertz unit of frequency to be detected functional requirement TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern ND number D number T ND number T D number ND number D number V ND number V ms ns us to be not detected functional requirement to be detected technical requirement to be not detected technical requirement to be detect
25. essary AS one example due to the requirement to detect all abnormal events it is difficult to avoid normal events from occasionally leading to detection In these cases the control should only be disabled temporarily and recover automatically since having to re enable the control manually is troublesome for the user if it must be done frequently On the other hand it is obvious to require a manual reset after the operator has fallen when the remote is left unused when it senses a hard impact as well as when continuous weightlessness is detected which indicates that the device is falling from a tall height These events are not supposed to occur during normal usage and can be considered critical Different from the original algorithm this one has to detect additional events besides falls while assuring as little overlap with normal behavior as possible see 3 1 Functional requirements However in order for the detection to be reliable every single abnormal situation must be detected This requires sufficiently high sensitivity which may cause even slight harmless movements to trigger an interrupt Since safety cannot be less important than usability it must be accepted that there are movements that will temporarily disable the control For instance when walking downstairs weightlessness may be detected occasionally In this case the algorithm will start over and re enable the control when the expected strike after weightlessness does not o
26. hat it shuts down see 3 1 Functional requirements When the device was assigned a free fall acceleration threshold of 0 3 g weightlessness was detected from no heights less than 10 cm With a threshold of 0 6 g it was detected till 5 cm 3 uses even 0 75 g which makes the detection very reliable but also increases the chance of confusion with other events as pointed out in 4 2 1 D2V Remote left unused This case was detected successfully However for the end use it may be necessary to adjust the time after which the remote is classified as unused D3V Remote receiving a shock This case was detected successfully TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 42 D4V Invalid position This case was detected successfully NDOV Operator jumping down As can be seen in NDOI this situation results in a restart of the algorithm after the resulting position is found to be equal to the initial position Therefore it is successfully not detected as abnormal behavior if the operator jumps down small heights However if he she jumps high or jumps down from a place higher than ca 40 cm this is identified as a continuous fall which is a critical event that must require a manual reset Either way the requirement of allowing the operator to jump down small heights such as 30 cm has been met ND1V Operator standing still or sitting with remote As explained in ND11 see 4 2 2 this case has not been implemented a
27. he only difference is that in this case the remote doesn t turn over which can be seen when comparing the acceleration pattern at the beginning and end of the graph TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 25 100 23 30 32 SM 36 23 25 2 39 3 J a So 400 900 K WV re ae 400 VALUE 256 9 1900 2400 2900 SAMPLES 50 s Graph 2 Acceleration change curves of remote control being dropped D2T The remote may be left unused This means it must be detected that the acceleration on each axis doesn t change D3T Remote may receive a shock A sudden high rise in acceleration is to be detected D4T The remote may be in an invalid position This requires detection of how much the acceleration for each axis differs from the initial state see Inclination position check D5T Losing balance Depending on the direction and intensity of stumbling or slipping the data for this case can have various shapes However in general a loss of balance starts with a sudden increase in acceleration the cause of the stumbling e g stepping TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 26 in a hole or collision with an object followed by a loss of balance If the operator ends up falling this is detected by DOT Below are examined the technical requirements for the situations that shall not cause detection NDOT VALUE 256 g The graph below shows the operat
28. icate a reaction to abnormal behavior which wasn t implemented such as disabling the remote possible reactions are examined in 4 2 1 Green letters indicate a restart of the algorithm OxFO i e the control is re enabled The resulting state machine is depicted in Figure 15 where bold green connectors indicate transitions that were added TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern Free fall interrupt disable Ctl Strike timed out enable Ctl Inactivity interrupt Figure 14 The relevant states of the original algorithm TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern Stable timed out enable Ctl Stable status equal to initial status enable Ctl 33 34 Activity interrupt Inactivity interrupt Inactivity interrupt disable Ct Inactivity interrupt Position invalid disable Ctl Position valid enable Ctl Free fall interrupt Free fall interrupt disable Ctl Strike timed out enable Ctl Inactivity interrupt Stable status equal to initial status enable Ctl Figure 15 The derived state machine When testing the implementation the detection statuses that can be seen in the state machine above are logged for later evaluation on the computer Additionally they re indicated at run time using LEDs and buzzers which provide immediate feedback so that the data doesn t need to be transmitted to the PC after every test Subsequently the
29. is of scenarios that can go wrong It is difficult to estimate how many ways there are to use a machine inadequately and if a machine is certified to comply with modern safety standards this doesn t guarantee safety Instead this increases the risk that people feel protected and therefore act carelessly This shows that safety can only be increased but never guaranteed Timo J skel inen introduced me to Technion Ltd subsequently referenced as Technion where he worked After an interview about my level of knowledge was offered a project for my bachelor thesis and accepted Technion is studying accelerometers in order to find out whether they re suitable to increase remote control safety Accelerometers measure acceleration They re used for instance in biology engineering and everywhere in our daily life including gaming and mobile phones 13 The remote is used to control vehicles as for instance excavators Due to inadequate usage or accidents it may receive uncontrolled input and the controlled machine could injure nearby humans or damage the environment Therefore the accelerometer is used to observe the status of the remote and detect abnormal behavior which requires stopping certain commands from being transmitted to the machine While increasing safety during remote control usage best possible usability should be preserved i e normal behavior such as standing walking and sitting with the remote should not interrup
30. le waiting for an impact 1 2 Next an invalid position is detected in the middle of the fall due to inclination 3 which doesn t prevent the valid fall sequence from being detected 4 7 This case was tested outside on grass field as well as inside on the floor padded with foam Due to the risk of injury falling could only be tested less TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 41 severe than it might be in a real situation Still the same results as in AN 1023 have been reached while falling to each side and even from small heights falling from squat position The result is satisfying especially in the context of other detections D1V Remote being dropped This safety feature was tested by dropping the remote into a cardboard box padded out with foam from different heights Either one of the two sequences mentioned in 4 2 2 D11 was always detected successfully When motionlessness is detected state OxF3 and the position after the fall is equal to the initial position the control can be re enabled Otherwise the system may shut down Note that in the current implementation this case isn t distinguished from the operator jumping down However this can be done as described in 4 2 2 D11 and will be necessary before using the application in the final remote where the user may not want to reactivate the system every time after jumping down from somewhere Dropping the remote however may require t
31. mmunicate with a machine The IO expander is used to support for instance joysticks LEDs and buttons For development and testing a prototype was used which doesn t include a radio controller or peripherals because the focus was not the communication with a machine but testing whether the ADXL345 is suitable to gather status information Picture 1 shows the real remote and the prototype that was used 11 1 1O expanders increase the amount of available IO pins TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 10 Picture 1 Technion TRC remote control left and the used prototype right 2 1 2 ADXL345 accelerometer The ADXL345 is a 3 axis accelerometer by Analog Devices which can measure acceleration of up to 16 times gravity of Earth which is expressed in g whereat 1 g equals approximately 9 81 m s varying slightly depending on altitude latitude and other factors in relation to the position of measurement 10 The part has an output resolution of up to 13 bit and measures both dynamic acceleration resulting from movement or shock and static acceleration from gravity The latter can be used to measure inclination whereat changes of less than 1 0 can be resolved But it isn t limited to pure acceleration data output Instead it offers special functions that can serve very specific applications see paragraph Functions The acceleration data can be read from the part s data registers see paragraph Data format
32. nd is thus misleadingly detected as remote left unused 5 2 Response time analysis Since the detection has to be sufficiently fast and since it will share CPU time with other tasks analysis of response times was required Subsequently the results are discussed Since the rise and fall time of the interrupt pin takes some ns it s negligible and the I C transmission time takes some us up to a few ms So only the time for reading and writing the part s registers must be examined The best case is when no interrupt needs to be processed and detect_behavior returns immediately which takes less than 1 ms If the function is executed and an interrupt has occurred the processing time depends on the I C transmissions Depending on the state transition up to 4 registers are written sequentially which takes approximately 23 ms During a position check 6 sequential registers are read which takes approximately 34 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 43 ms FreeRTOS function xTaskGetTickCount was used to measure the times see 2 2 1 FreeRTOS Since the time between the occurrence of an interrupt and the beginning of its processing which is the start of function detect_behavior varies between 0 ms and task delay the following can be concluded about the time from the occurence of an interrupt until it has been served The time is 0 task delay processing time of detect_behavior Best ca
33. nd it is desirable to have the control disabled as soon as possible Therefore the following actions are reasonable after contact with the ground has been detected The time parameter for the Inactivity interrupt is decreased so that an unused remote is detected sooner It is increased again after a change of state Atthe same time the acceleration threshold is decreased so that a user who is moving on or standing still after jumping down doesn t cause the Inactivity interrupt only the remote on the ground shall cause it Additionally a timer is necessary while waiting for motionlessness If it times out there has been movement after the fall and it is known that the operator has jumped down In this case the algorithm proceeds with the initial status TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 36 That means the sequence F1 gt F2 gt F3 gt F0 is split up into F1 gt F2 gt F3Inactivity gt final state for a remote fall and F1 gt F2F3 timeout while waiting for inactivity gt FO if the carrier jumped down When dropping the remote from taller heights the following sequence should be detected Different from the above sequence this one is acceptable because both jumping down and falling from high places are dangerous and must therefore lead to detection 1 Weightlessness status OxF1 2 Continuous free fall critical OxFF D2 Remote left unused This situation is detected by the Inactivity interr
34. of a reading operation the acknowledge bit is directly followed by a data byte from the slave If it is a TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 19 writing operation the master transmits a data byte As long as the master doesn t transmit the stop condition further bytes of data can be transmitted 5 The process of writing and reading via 1 C is depicted in Figure 6 and Figure 7 ee A ee GS data transferred ol as in bytas acknowledge LI from master to slave A acknowledge SDA LOW A nol acknowledge SDA HIGH g from slave to master START condition F STOF condilion Figure 6 Writing n bytes using I C protocol 5 S stave nooness oe cate PL dala transherred a in byles acknowledge Figure 7 Reading n bytes using I C protocol 5 The LPC1768 microcontroller is interconnected with the ADXL345 via 12C whereat the microcontroller functions as the master and the accelerometer as the slave see 2 3 12C The used I C operating speed is 100 kHz 2 4 Existing work 2 4 1 Application Note 1023 Fall detection Analog Devices provides application notes for the ADXL345 among those is an essay where the part is belt wired AN 1023 The application is based on the experience that for senior citizens the biggest risk about falling is the delay or TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 20 absence of help after the accident It has been used an
35. ood to use a successor model of the ADXL345 The ADXL346 offers some enhancements whereat the main difference is lower power consumption REFERENCES 1 A Brief History And Overview Of The Accelerometer Online article Available from http Awww beembee com 201 1 a brief history and overview of the accelerometer cited 6 2012 2 ADXL345 data sheet Rev C PDF document Available from http Awww analog com ADXL345 cited 6 2012 3 Application Notes 1023 PDF documents Available from http www analog com en mems sensors mems inertial sensors adx 345 products technical documentation index html cited 6 2012 4 Application Notes 1077 PDF documents Available from http www analog com en mems sensors mems inertial sensors adx 345 products technical documentation index html cited 6 2012 5 UM10204 Il2C Bus Specification and User Manual NXP Rev 03 19 June 2007 Available from http Awww nxp com documents user_manual UM10204 pdf cited 6 2012 6 Accelerometer Wikipedia article Available from http en wikipedia org wiki Accelerometer cited 6 2012 7 l C Wikipedia article Available from http en wikipedia org wiki l2c cited 6 2012 8 Analog Devices support forum Available from http ez analog com welcome cited 6 2012 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 46 9 ADXL345 Question about reference values for AC coupled mode Forum thread Available from http e
36. or jumping down from a height of 30 cm holding the remote still to prevent it from bouncing The similarity to dropping the remote D1T is visible so a characteristic has to be found in order to distinguish them 400 200 0 latr E P mer 3 7 AT 15 19 23 a39 43 47 51 55 50 GB 67 71 75 79 83 87 91 14 9 4 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 39 200 400 X y 600 z 800 1000 1200 SAMPLES 50 5 Graph 3 Jumping down from chair ND1T Standing still and sitting with the remote control on the lap are basically equivalent Since the remote is in use this situation TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 27 should not be detected as abnormal The graph below shows that there is no significant element except for a constant tremor in the acceleration Since the magnitude of the tremor is rather small it may be difficult to determine that the remote is not left unused 300 250 m 200 S to 150 x X 100 Q 3 ai 50 gt z 11 ST 51 71 91 111131151 171 191 211 231 251 271291 311331351371 391 50 4 21 44 614 814 101 121 141161 181 201 221 241 261 281 301 321341361 381 Samples 50 s Graph 4 Standing still with remote 4 REALIZATION An existing project in C language which uses FreeRTOS see FreeRTOS was used as the underlying code base and an ADXL345 driver was designed that uses an underlying lC driver Throughout the project the ADXL345 was used with an output
37. ponse vs Orientation to Gravity ADXL345 data sheet 12 Figure 3 Format of data registers in 13 bit mode AN 1077 c c seeeeeeeeeeeeeeeeteees 12 Figure 4 Minimum Initialization SEQUENCE eee eeeeceeeceeee eee eeeenenaeeeeeeeeeeteneenaaeaes 17 Figure 5 FG connection diagram 2 siceicicecdoseatcetas hie helenae eae eee delete 18 Figure 6 Writing n bytes using C protocol 5 cee eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeteeeenaaees 19 Figure 7 Reading n bytes using I C protocol 5 ccccccceeeeeeeeeeeeeeeeeeeeeeeeeeeesaaeees 19 Figure 8 Flow chart of the original algorithm 3 cc ceeeeeeeeeeeeeeeeneeeeeeeeeeeteeeeaaees 21 Figure 9 Code structure of the original algorithm c ccceeeeeeeeeeeeeeeeeeeeeeeeeeeennaaees 22 Figure 10 Role of timer variables in detect_behavir ccccceeeeeeeteeeeeeeeeeeeeeeeeeees 22 Figure 11 Structure of the function derived from the original interrupt handler 31 Figure 12 Correlation between task delay and time WINdOWS ccceeeeteeeeeteees 32 Figure 13 DELAY_XL is used both as the task delay and to calculate correct values for the time intervals of the algorithm x c seies ete tive cence cdente a ctiee nets detecerareiend Aeterna 32 Figure 14 The relevant states of the original algorithim ceeeeeeeseeeeeeeeeeeeeeeeteees 33 Figure 15 The derived state MACHING ccccceeeeeeeeeeenceeeeeeee
38. s impact with the ground and stability have been detected the algorithm compares the difference in acceleration on each axis to the initial status If the difference exceeds a certain threshold this means the user has turned over and the fall is valid This is the only opportunity where acceleration TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 23 data is processed by the application Every other function uses only the part s interrupts 2 4 2 Competitor solution HBC radiomatic has implemented similar remote control safety features 15 3 REQUIREMENTS Consecutively the requirements for this project are listed 3 1 Functional requirements The functional requirements defined the following use cases as abnormal situations which is why they should cause detection DO The operator can fall down to any side left right front back A fall is only valid when the operator ends up lying on the ground D1 The remote control can be dropped D2 The remote may be left unused D3 The remote may receive a shock D4 The remote may be in a position that doesn t allow for adequate usage D5 The operator may lose his balance e g due to a slippery surface or because he steps into a hole The following situations were to be distinguished from the above cases and must not cause detection The reason is that in the final application the user must be able to move around freely while using the remote to control a machine
39. se An interrupt e g the Activity interrupt occurs right before the task delay ends The response time is ca 0 23 ms 23 ms Worst case The position check timer indicates that it s time for a position check and additionally an inactivity interrupt occurs after a strike has been detected status OxF2 which leads to another position check Furthermore assume that the inactivity interrupt has occurred right after the task was delayed with vlaskDelayUntil The response time is ca task delay 2 time for position check task delay 68 ms Note that these are only the response times for the serving of one interrupt A fall event for instance involves multiple sequential interrupts in order to be classified as a fall 5 3 Personal learning experience This project was very enriching On the one hand it was incredibly valuable to gather an understanding of the commercial context of such a project It required several iterations during the work process in order to find out which factors are relevant and which are rather a waste of time On the other hand it was challenging to abstract from the elements in the work process in order to build a logical thesis text TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 44 5 4 Employer s feedback According to Technion the goals of the project were met and the resulting data samples code and documentation as well as this thesis are useful for future work 5
40. t the transmission of commands The goal of the project is to build up a knowledge base related to the usage of accelerometers and their role in the detection of above mentioned behavior Furthermore code which provides the functionality for the above has to be produced The resulting code is not ready to be used in the final remote product Its main purpose is to demonstrate what is possible with an accelerometer and to examine whether it suits the requirements TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern Chapter 2 gives an overview of the used hardware and software Chapter 3 discusses the requirements Chapter 4 describes the implementation of the solution points out difficulties and gives reasons for the decisions that have been made during the work process Chapter 5 discusses whether the requirements have been met validates the applicability of the solution for the final product and proposes improvements for the future 2 BACKGROUND INFORMATION This chapter describes the hardware and software that has been used and gives an overview over existing similar work see 2 4 Note that only the most important background information is described For details see the ADXL345 data sheet and the application notes by Analog Devices 2 3 4 2 1 Hardware 2 1 1 The remote control The remote control is designed by Technion and includes an IO expander in combination with a microprocessor and a radio controller to co
41. upt The parameters are the inactivity time and the acceleration threshold It may seem obvious to exit the state left unused when activity is detected just like it was entered due to the detection of inactivity However an activity interrupt may not come from the operator but from the environment and must not lead to re activation Therefore left unused is a final state that may require manual reset of the control D3I Remote receiving a shock The detection of this event starts with the Activity interrupt which leads to state OxF6 A shock against the remote is not only detected as an impact after a fall but also within the initial state Still the detection process is the same as during a fall after weightlessness has been detected and includes the position comparison at the end The reason for separating Strike after weightlessness OxF2 and Strike Shock OxF6 is that the activity acceleration thresholds are different in both cases The reason for separating the successive stable statuses OxF3 and OxF7 are different thresholds for the inactivity interrupt in TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 37 both cases complete motionlessness after a fall vs significantly less movement than during a shock strike D4I Invalid position This detection uses the position check from the original algorithm Instead of only using it during a fall event it now constantly monitors the remote s position too If it is
42. used Tasks are created using function xTaskCreate The task is assigned a priority which needn t be unique However the idle task has priority 0 FreeRTOS allows for both preemptive and co operative scheduling 14 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern 18 2 2 2 Compiler Sourcery G Lite ARM EABI This is CodeSourcery s version of the GNU toolchain It includes compilers assemblers linkers and libraries The version used in this project is 2009q1 161 2 2 3 Serial communication PUTTY PuTTY is an SSH and telnet client that has been used for serial communication with the microprocessor especially for logging of axis data 2 3 C serial bus protocol l2C Inter integrated circuit is a multi master single ended serial bus standard that uses only 2 wires between master and slave SDA serial data signal is used by the host processor master for reading and writing and SCL serial clock is set by the processor It supports 7 bit as well as 10 bit addressing 2 Vop vo Figure 5 l C connection diagram 2 Transmission protocol Each transmission starts with a start signal and ends with the stop signal by the master The start command is followed by a byte with the 7 bit slave address and the R W bit which indicates whether the master wants to read or to write Next the slave transmits an acknowledge bit to signal the transmitter master that a byte was successfully received In case
43. z analog com message 51880 51880 cited 6 2012 10 Gravity of Earth Wikipedia article Available from http en wikipedia org wiki Gravity_of_Earth cited 6 2012 11 Technion Wireless controllers Website Available from http technion fi wireless_controllers html cited 6 2012 12 Graham B B 2000 Using an Accelerometer Sensor to measure Human Hand Motion Thesis Available from http www mtl mit edu researchgroups MEngTP Graham_Thesis pdf cited 6 2012 13 Noton A 2011 A Brief History And Overview Of The Accelerometer Article Available from http Wwww beembee com 201 1 a brief history and overview of the accelerometer cited 6 2012 14 FreeRTOS Quick Start Guide Website Available from http www freertos org FreeRT OS quick start guide html cited 6 2012 15 Remote control safety features Website Available from http www hbc radiomatic com en Products Special functions features Details items Safety features html cited 6 2012 16 Application Notes 1025 PDF documents Available from http www analog com en mems sensors mems inertial sensors adx 345 products technical documentation index html cited 6 2012 TURKU UNIVERSITY OF APPLIED SCIENCES THESIS Lukas Kern
Download Pdf Manuals
Related Search