Wire Less Sensors. Getting more out of what you have, or doing the same job with less. Laboratory for Electromagnetic and Electronic Systems MIT

Transcription

1 Wire Less Sensors Getting more out of what you have, or doing the same job with less. Steven B. Leeb and Team Laboratory for Electromagnetic and Electronic Systems MIT

2 Buildings Today Centralized or no control Haphazard monitoring Invasive wiring and wireless connections

3 No Watt Left Behind Distribute control Deploy central monitoring Signal processing for diagnostics/prognostics Preserve data privacy

4 Many cut-off ears

5 Many cut-off ears But do we have any Van Goghs? (with thanks to Jim Williams)

6 Non-Intrusive Sensing

7 Transient-Based Identification Transient electrical behavior is strongly influenced by the physical task performed dby the load

8 Spectral Envelopes Stator current and P 1 for a 3-phase induction motor

9 Upstream and Downstream

10 Spectral Envelopes

11 Applications Works for any system with a power distribution network (AC or DC):

12 Connections with Massachusetts School Building Authority Field tests to be conducted at sites selected in collaboration with MSBA. Additional opportunity for educational outreach.

13 USCG Cutters Seneca and Escanaba Field Experiments USN DDG 51 USN DDG-51 LBES

14 Ship-board Installation

15 NILM on a Microgrid One device per panel is collecting measurements from the microgrid serving 150 people.

16 One-line Diagram

17 Electrical Sensing

18 Non-Contact Current Sensor local fields are non-zero

19 Noncontact current/voltage sensing

20 TMR current sensor

21 TMR Feedback

22 Capacitive Voltage Sensor

23 Differential Voltage Sense

24 Easy Installation

25 WaterWOLF Turning a standard water meter into a high bandwidth flow monitor

26 WaterWOLF Internal fly wheel couples magnetically to external metering device.

27 WaterWOLF Two compensated TMR sensors Wt WOLF t fit t t d d t i t WaterWOLF retrofit converts any standard meter into a real time, high bandwidth water monitor.

28 Global Access and Security

29 Solving the Big Data Problem

30

31

32

33

34

35 Energy Scorekeeping

36 48 hours of Electrical Data Panel 1 Panel 2 Phase A Phase B Phase C Phase 2A Phase 2B Phase 2C Watts Watts Zoom In Friday Saturday Sunday Q: How do we make this data useful?

37 1 hour of electrical data How can we be sure these are actually heater and fridge events? Zoom in Zoom in further Phase A Phase 2A Phase B Phase C Heater On Heater Off Phase 2B Phase 2C Heater On Fridge On Fridge Off A: By identifying every electrical event

38 8 seconds of electrical data Software can find common features: 1. Always the same peak value 2. Always the same steady-state state change 3. Always the same start-up duration 4. Always the same phase 5. Always the same signature Zoom in further Phase 2A Phase 2B Phase 2C peak start up duration steady state change

39 Digital Electronics Macbook Pro as seen by the NILM Charger plugged in Shutdown Laptop plugged in and boots High CPU load Idle Charging

40 Cross Corroboration Comparing existing sub-meter data from every tent/structure over the same 48 hour period: Ft. Devens sub-meters NILM Estimate 4876 kwh 4852 kwh Major loads accounted for over 99% of total consumption Note: NILM estimate =

41 Activity tracking and scheduling verification

42 Power Consumption (kwh) during the training weekend Extracted from 4876 kwh Total Represents difference bt between total t use and major load use * 48 hour time period, Nov 2013

43 48 hours of Electrical Data Panel 1 Panel 2 Phase A Phase B Phase C Phase 2A Phase 2B Phase 2C Watts Watts Friday Saturday Sunday

44 Equipment Hours during the training weekend Extracted from * 48 hour time period, 4876 kwh consumed

45 Pump Events Good indicator of Human Activity events No pump events Unit departed Sleeping after events (compared to 8 the previous evening and 13 the previous morning) probably showered No events Sleeping events Events at 0437, 0556, and events Unit returned Wake up time 0600 morning hygiene before events between Three pump events 10 occur before 6:33 Unit scheduled to arrive at 1600 Only 1 event while at range Main i body probably bbl lfb left by Immediately used the bathrooms personnel stayed behind 0900 Note: 4 identical pumps (2 latrine, 2 shower), each with gallon septic tank

46 48 hours of Electrical Data during a training weekend What they actually used: 913 kwh Panel 1 Unoccupied use (different day, similar temperature): 209 kwh Panel 2 Phase A Phase B Phase C Difference of 704 kwh 14% of total Phase 2A Phase 2B Phase 2C Watts Watts Friday Saturday Sunday Q: What if unit turned dthings off while at the range?

47 Perform Condition-Based Maintenance

48 Phase A Phase B Phase C 1 Operation of the ECU (Cool Mode) Power (Watts) loaded unloaded loaded unloaded 2 Time (Hours : Minutes : Seconds) 15s period Sequence of operation 1. Supply fan on 2. 3 second delay 3. Condenser fan on 4. Compressor turns on using soft start controller 5. Compressor (two states, loaded and unloaded) Over a 15 second period, compressor is loaded 10% of the time

49 Fault Identification thru NILM (ECU Cool Mode) Compressor Restarting Too Frequently Phase A Phase B Phase C Zoom in Explanation: Every 5 10 minutes, this same compressor starts, runs for a short time, and turns off. It could be malfunctioning, or perhaps the A/C is just not needed (space is cool already) and should be turned off. Software code can look for common faults like this and alert the base manager.

50 Back-up flow meter When septic tanks fill, electric pumps activate Each pump has it own gallon tank Pump cycle = 12 gallons used Example: 27 (8AM) 29 (10AM) September BCIL was occupied flow meter measured 1150 gallons used NILM counted 46pumpevents (monitoring2 of 4 pumps ) NILM counted 46 pump events (monitoring 2 of 4 pumps ) ~max 552 gallons (times 2 = 1104 gallons, decent estimate!)

51 Cycling System Faults Examples: Waste-disposal systems LP air systems Leaks are frequent Leaks and high usage have the same effect on the actuator Example: SENECA s waste disposal system

52 System Overview Control Panel NILM Pressure Switch Level Indicating Probes/ Discharge Pump Control VCT Alarm Vacuum Pumps Pump On Pump Off Sewage Discharge Pumps

53 Pump Operation

54 Click to edit Master title style 54

55 Sensor Line Clog Clogged pressure switch gauge line VCT orifice before (left) and after (right) cleaning in response to a casualty.

56 Clog Signature Normal three-hourhour power plot (top) Pressure switch casualty power plot (bottom).

57 Clog Signature

58 When nonintrusive is tough.

59 VAMPIRE Power Electronics

60 VAMPIRE Vibration Assessment Monitoring Point with Integrated Recovery of Energy Equipment to be Monitored Motives Minimally intrusive No battery & No special power wiring No data wiring: wireless communication Service-free after the installation

61 Data logger (up to 32 Gb) 140 g mass, affixed rated to 4kg 3 axis acceleration at 3.2 khz bandwidth Vibration Monitoring

62 Machine Imbalance for Virtual Vibe Input Rotating Imbalance: Amplitude proportional p to speed squared Freq equal to shaft speed Sweeps Vibration Mounting Resonance 62

63 Mounting System Identification Virtual Vibration Input from Spin-down Measured Vibration Output

64 H1 FRF Estimator Y(f) = X(f)H(f) + N(f) H1(f) = G(XY)(f)./G(XX)(f) Virtual Vibration i Input from Spin down: x(t) () Measured Vibration Output: y(t) Noise/uncorrelated signal from environment: n(t) Figure Credit: Noise and Vibration Analysis: Signal Analysis and Experimental Procedures by Anders Brandt

65 Mounts and Error Check Simulated Mount Condition i FRF () (a) Check for Steady Environment Error With Delayed Signal FRF (b)

66 USCGC SENECA

67 Experiment USCGC SENECA Vent Fan

68 Mount Stiffening

69 Fan Imbalance Wire Bundle

70 SENECA Fan FRFs

71 SENECA Fan FRFs Increasing Stiffness and Resonance Frequency

72 Wire less sensors An opportunity to look at your system or An opportunity to look at your system or product and squeeze it for more