EP4639097A1 - Recalibrating sensor values - Google Patents

Recalibrating sensor values

Info

Publication number
EP4639097A1
EP4639097A1 EP23818020.2A EP23818020A EP4639097A1 EP 4639097 A1 EP4639097 A1 EP 4639097A1 EP 23818020 A EP23818020 A EP 23818020A EP 4639097 A1 EP4639097 A1 EP 4639097A1
Authority
EP
European Patent Office
Prior art keywords
data stream
sensor unit
sensor data
calibration
sensor
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
EP23818020.2A
Other languages
German (de)
French (fr)
Inventor
Ian Thurlow
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
British Telecommunications PLC
Original Assignee
British Telecommunications PLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from GBGB2219548.1A external-priority patent/GB202219548D0/en
Application filed by British Telecommunications PLC filed Critical British Telecommunications PLC
Publication of EP4639097A1 publication Critical patent/EP4639097A1/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
    • G01D18/00—Testing or calibrating apparatus or arrangements provided for in groups G01D1/00 - G01D15/00
    • G01D18/002—Automatic recalibration

Definitions

  • the present disclosure relates to recalibrating sensor values using a mobile reference sensor unit.
  • Electronic sensors are widely used to produce data representative of a physical phenomenon such as temperature, humidity etc. Generally, they produce an output that has a derived relationship with the physical phenomenon that can be used to convert an output signal of the electronic sensor into data.
  • the output is the product of the interaction between an electronic circuit in the sensor and its surroundings.
  • Certain components in the electronic circuits are susceptible to change their output when factors in their surroundings change. For example, amplifiers can be prone to output drift over time particularly when there is a variation in temperature. Their inclusion in a sensor means that, inevitably, the output will drift away from the derived relationship with the physical phenomena, leading to errors.
  • Sensors with high quality components such as low-drift amplifiers are expensive and do not eliminate drift completely.
  • Sensor drift can be compensated for by calibrating the sensor using a separate control sensor that is known to be correct.
  • sensors could be adjusted following calibration by physically changing the circuit to compensate for the drift.
  • One example is by adjusting a variable resistor until the output is changed closer to the expected value. This type of manual adjustment is cumbersome, time consuming, and expensive. Further, it cannot be applied retroactively to measurements already taken.
  • a computer-implemented method of recalibrating sensor values using a mobile reference sensor unit comprising: storing a raw sensor data stream received from a sensor unit; receiving a first set of calibration measurements from the mobile reference sensor unit over a first calibration period; generating a calibrated sensor data stream from the raw sensor data stream using the first set of calibration measurements; receiving a second set of calibration measurements from the mobile reference sensor unit over a second calibration period later than the first calibration period; generating a second calibrated sensor data stream from the raw sensor data stream using the second set of calibration measurements; identifying a time between the first and second calibration periods where there was a change of drift in the sensor unit using the raw sensor data stream and the first and second sets of calibration measurements; and generating a recalibrated sensor data stream comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.
  • the method scales to deployments of thousands or tens or thousands of sensors units such as in smart cities where the sensor units are on streetlamps or in other scenarios.
  • Efficiency is gained by using a mobile reference sensor unit so that it is possible to have fewer mobile reference sensor units than sensor units.
  • the mobile reference sensor units are able to move about to different ones of the sensor units giving scalability. Since the mobile reference sensor units may have more expensive or more sophisticated sensors than the sensor units, efficiency is gained (as there are fewer mobile reference sensor units). Also the sensor units themselves may be more vulnerable than the mobile reference sensor units.
  • the sensor unit and the mobile sensor are both one of: air temperature, air quality, relative humidity, wind speed and road surface temperature sensors.
  • air temperature air quality
  • relative humidity relative humidity
  • wind speed road surface temperature sensors
  • the sensor unit is at a fixed location. Where the sensor unit is at a fixed location there is a particular problem of scalability, since as the number of sensor units increases so does the geographical area of the site where the sensor units are deployed.
  • the sensor unit is periodically relocated and optionally generating the recalibrated sensor data stream is carried out only up to a time when the sensor unit was most recently relocated.
  • Being able to periodically relocate the sensor unit brings flexibility, such as where a sensor on a portable street lamp is used. The process is able to operate well even where the sensor unit is periodically relocated. If a sensor unit is relocated, say, from an area where the air quality is good (e.g. low levels of particulates) to an area where the air quality is poor, there is likely to be a break in a time series of the sensor data stream (during the period of the relocation) followed by a significant jump in the values of the sensor data stream. In this case any recalibration is applied only back to the date/time the sensor unit started to resend data from its new location.
  • the air quality e.g. low levels of particulates
  • the mobile reference sensor unit is attached to a vehicle which is any of: an unmanned vehicle or an unmanned aerial UAV.
  • a vehicle which is any of: an unmanned vehicle or an unmanned aerial UAV.
  • At least one of the first and second set of calibration measurements is collected within a threshold distance of the sensor unit.
  • Improve accuracy may be achieved as a result since the calibration measurements generally have less noise where the sensor unit and mobile reference sensors are within a threshold distance of one another.
  • the mobile reference sensor unit provides more accurate values than the sensor unit. Where the mobile reference sensor unit is more accurate its measurements may be used for calibration with good results.
  • the mobile reference sensor unit is calibrated more frequently than the sensor unit. By calibrating the mobile reference sensor unit more frequently, it is possible to have better accuracy at the mobile reference sensor unit than at the sensor unit.
  • identifying the time further comprises: generating a first drift prediction for the raw sensor data stream using the first set of calibration measurements; generating a second drift prediction for the raw sensor data stream using the second set of calibration measurements; and identifying the earliest time between the first and second calibration periods when the second drift prediction fits better to the raw sensor data stream than the first drift prediction. This is an efficient and effective way to compute the identified time.
  • the first and second drift predictions are generated using any one or more of: linear regression, extrapolation, using a machine learning model, using another artificial intelligence model.
  • the drift predictions are computed quickly and accurately in a way which is scalable.
  • the method is performed by a remote server, and wherein the first and second calibration measurements are received through any of: a wireless communication, a wireless communication with higher quality than a communication link between the sensor unit and the remote server.
  • a wireless communication a wireless communication with higher quality than a communication link between the sensor unit and the remote server.
  • the method comprises receiving any of the first and second calibration measurements from a store at the mobile reference sensor unit, in response to a communication link with the mobile reference sensor unit becoming available.
  • the method is robust since calibration measurements are unlikely to be dropped in transmission between the mobile reference sensor unit and the remote server.
  • the communication link with the mobile reference sensor unit becomes available as a result of a wireless communication link becoming available or as a result of a fixed network connection becoming available in response to the mobile reference sensor unit docking into a unit with a fixed network connection.
  • the calibration data is effectively and efficiently sent to the remote server module.
  • a computing device comprising: one or more processor; and a memory comprising instructions that when executed by the one or more processors, cause the processor to: store a raw sensor data stream received from a sensor in the memory; receive a first set of calibration measurements from the mobile reference sensor unit at a first calibration time; generate a calibrated sensor data stream from the raw sensor data stream using the first set of calibration measurements; receive a second set of calibration measurements from the mobile reference sensor unit at a second calibration time later than the first calibration time; generate a second calibrated sensor data stream from the raw sensor data stream using the second set of calibration measurements; identify a time between the first and second calibration times where there was a change of drift in the sensor unit using the raw sensor data stream and the first and second sets of calibration measurements; and generate a recalibrated sensor data stream comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.
  • a system comprising: a sensor unit; a mobile reference sensor unit; and a computing device as described above.
  • FIGs. 1A & 1 B are schematic diagrams illustrating a mobile reference sensor unit and a sensor unit communicating sensor measurements to a remote server through a communication network;
  • FIG. 2 is a flow diagram illustrating steps for recalibrating a sensor data stream using a mobile reference sensor
  • FIGs. 3A & 3B are graphs illustrating the calibration of a raw sensor data stream
  • FIG. 4A is a graph illustrating an example identification of a time where sensor drift has changed
  • FIG. 4B is a graph illustrating a recalibrated sensor data stream
  • FIG. 5 is a flow diagram illustrating part of FIG. 2 in greater detail
  • FIG. 6 is a schematic diagram of a remote server module.
  • FIGs. 1 A and 1 B of the accompanying drawings show a system 100 comprising a remote server 102, a mobile reference sensor unit 104 and a sensor unit 106.
  • the mobile reference sensor unit 104 and sensor unit 106 are both communicatively connected to the remote server 102 by way of a communications network 110 such as the internet, an intranet or any other communications network .
  • the sensor unit 106 includes one or more sensors suitable to take measurements from an area surrounding the sensor unit 106.
  • the sensor unit 106 also includes network equipment suitable for packetizing and sending the measurements to the remote server 102.
  • the sensor unit 106 repeatedly measures at least one physical quantity from the surroundings and sends the measurements to the remote server 102 through the communications network 110.
  • the sensors included in the sensor unit 106 are one or more of: air temperature sensor, air quality sensor, humidity sensor, wind speed sensor, road surface temperature sensor.
  • the remote server is a web server, a compute node of a communications network, or any other computing entity able to receive measurement data from the mobile reference sensor unit 104 and the sensor unit 106.
  • the sensor unit 106 is positioned at a desired site where it is desirable for the one or more sensors of the sensor unit 106 to measure the physical quantities of the surroundings.
  • the sensor unit 106 is shown fixed to a streetlight in the example of FIG. 1A which shows a system 100 for recalibrating sensor data of the sensor unit 106.
  • FIG. 1A is an example and is not intended to be limiting as the sensor unit 106 may be located at any location that is accessible by the mobile reference sensor unit 104. In some examples, the sensor unit 106 is moveable allowing periodic relocation to different measurement sites.
  • Measurements are taken by the sensor unit 106 periodically. In some examples that means measurements are taken at regular time intervals measured in seconds, minutes, hours etc. In an alternative example the measurements are taken at irregular time intervals. Irregular time intervals are preferable where the physical quantity is expected to vary more during certain periods than others, factors affecting this include night/day, seasonal variations, traffic levels etc.
  • FIG. 1A and FIG. 1B each show a single sensor unit 106. However, in practice there may be thousands or tens of thousands of sensor units such as in a smart city, or wider metropolitan area or across a county.
  • measurements from the sensor unit 106 are received by the remote server 102 as a raw sensor data stream and are optionally stored at the remote server 102 or another location.
  • the raw sensor data stream includes for each measurement a measurement value and measurement time.
  • the raw sensor data stream is stored in time order; that is, the sensor data stream is a time series.
  • the measurement value is the output from the one or more sensors for example a voltage or current value.
  • the measurement value is converted at the sensor unit 106 into data expressing the physical quantity, for example degrees Celsius for air or road temperature.
  • the inventors have recognized that, over time errors increase in the raw sensor data stream because of sensor drift in the one or more sensors in the sensor unit 106. As discussed above, this is an inevitable consequence of the components found in the electric circuits of the one or more sensors. The time variant nature of sensor drift means that the error size between the true value and the measurements will change over time. Consequently, the stored raw sensor data stream at the remote server 102 has a variable error across time. To address this the inventors have developed a way of using a mobile reference sensor unit 104 to collect measurements separately from the sensor unit 106. The separately collected measurements from the mobile reference sensor unit 104 are used for calibrating and recalibrating the raw sensor data stream of the sensor unit 106.
  • the term “calibration” is used to refer to correcting error in sensor data measurements.
  • the term “recalibration” is used to refer to correcting error in sensor data measurements which have already been calibrated at least once.
  • the mobile reference sensor unit 104 comprises one or more sensors.
  • the one or more sensors comprise at least one sensor that measures a same physical quantity as the sensor unit 106.
  • the mobile reference sensor unit 104 comprises sensors that are expected to produce smaller errors than those in the sensor unit 106. In one example this is achieved by using higher quality sensors in the mobile reference sensor unit 106 than in the sensor unit 106. Alternatively or additionally this is achieved by more frequently calibrating the one or more sensors of the mobile reference sensor unit 104.
  • the mobile reference sensor unit 104 also comprises network equipment capable of sending calibration measurements to the remote server 102 through the communications network 110. In some cases the mobile reference sensor unit 104 has a memory to store the calibration measurements. When the mobile reference sensor unit 104 moves to a location where communication with the remote server 102 is possible, the calibration measurements are sent to the remote server 102.
  • the mobile reference sensor unit 104 is attached to a vehicle or other movable entity such as a temporary tower.
  • the mobile reference sensor unit 104 is attached to a wheeled vehicle which can be manned or unmanned.
  • the mobile reference sensor unit 104 is an aerial vehicle, such as an unmanned aerial vehicle (UAV).
  • UAV unmanned aerial vehicle
  • the use of a UAV produces a more efficient system 100 because it allows access to the site where the sensor unit 106 is located even when the site is inaccessible from the ground.
  • the mobile reference sensor unit 104 moves to visit the site of the sensor unit 106.
  • the mobile reference sensor unit 104 moves to the sensor unit 106 in any suitable way, such as by moving to a known location following instructions from the remote server 102.
  • the sensor unit 106 emits a beacon signal and when the mobile reference sensor unit 104 detects the beacon signal it moves towards the sensor unit 106.
  • the mobile reference sensor unit 104 begins to measure at least one of the same physical quantities as measured by the sensor unit 106.
  • the mobile reference sensor unit 104 is moved within a threshold distance of the sensor unit 106 based on the physical quantity being measured.
  • the threshold distance is around 1 to 2 metres in some cases.
  • the mobile reference sensor unit 104 moves within two metres of the sensor unit 106 before beginning to take measurements.
  • the mobile reference sensor unit 104 measures the physical quantity over the course of a calibration period.
  • the calibration period being set to a length such that it allows the mobile reference sensor unit 104 to record a plurality of calibration measurements that correspond to measurements taken by the sensor unit 106.
  • the remote server 102 receives both the raw sensor data stream from the sensor unit 106 and the plurality of calibration measurements from the mobile reference sensor unit 104.
  • both the raw sensor data stream and the plurality of calibration measurements are received through a wireless link via communication network 110.
  • the plurality of calibration measurements are received through a wireless link of higher quality than that used to send the raw sensor data stream to the remote server 102.
  • the mobile reference sensor unit 104 stores the calibration measurements until a connection to the communications network 110 with the remote server 102 becomes available.
  • the mobile reference sensor unit 104 is configured to take the measurements whilst physically proximate to the sensor unit 106, store the measurements in it’s memory, move to another location and dock into a unit that has a fixed wired connection to the communications network 110.
  • the fixed wired connection may be of higher quality than the communications link between the sensor unit 106 and the remote server 102. In this way the fixed sensor unit 102 may be low cost such as with lower quality data transmission equipment.
  • the remote server 102 comprises a memory configured to store the received raw sensor data stream and calibration measurements.
  • the remote server 102 uses a first set of calibration measurements to calibrate the raw sensor data stream stored from the sensor unit 106. As discussed in more detail below this first calibration eventually introduces further error by attempting to compensate for a level of drift that has since changed. Recalibration is therefore done. To carry out recalibration, the visit of the mobile reference sensor unit 104 to the site of sensor unit 106, shown in FIGs. 1A and 1 B, is repeated. Having received a second set of calibration measurements the remote server 102 recalibrates the raw sensor data stream. This recalibration is described in more detail regarding FIG. 2 below.
  • FIG. 2 illustrates a method 200 for recalibrating sensor values using a mobile reference sensor unit 104 such as the one described above in system 100.
  • the method is performed at a remote server such as the remote server 102.
  • a raw sensor data stream is stored having been received from a sensor unit such as the sensor unit 106.
  • Each measurement in the raw sensor data stream includes a measurement value and measurement time. The time between each measurement value is regular or irregularly spaced.
  • the measurement values represent values of a physical quantity measured by one or more of the sensors in the sensor unit 106.
  • a first set of calibration measurements is received.
  • the first set of calibration measurements having been collected during a first visit of the mobile reference sensor unit 104 to the site of the sensor unit 106.
  • the first set of calibration measurements is sent from the sensor unit 106 to the remote server 102 over the communications network 110.
  • Each measurement in the first set of calibration measurements includes a measurement value and a measurement time recorded by the mobile reference sensor unit 104.
  • the measurement values in the first set of calibration measurements represent the same physical quantity as those of the raw sensor data stream.
  • a first calibrated data stream is generated using the first set of calibration measurements.
  • the first calibrated sensor data stream includes the measurement values of the raw sensor data stream adjusted by the remote server 102, based on the first set of calibration measurements.
  • the amount of adjustment applied to the measurement values in the raw sensor data stream is based on an error between the first set of calibration measurements and corresponding values of the raw sensor data stream.
  • the error is computed in any suitable way such as by computing an average difference between the corresponding values in a specified time window, computing a mean squared difference between the corresponding values, computing a median difference between the corresponding values, computing a mode of the differences between the corresponding values or in other ways.
  • the first set of calibration measurements are determined by the remote server 102 to be larger than the measurement values for the raw sensor data stream at the nearest times then an offset adjustment is made to increase the values for all raw sensor data stream values.
  • Historic and subsequent measurement values for the raw sensor data stream received from the sensor unit 106 are then also adjusted and included in the first calibrated data stream. The measurement time values are not adjusted.
  • Calibrating the raw sensor data stream to produce the first calibrated sensor data stream in this way is more efficient and less expensive than on-site physical calibration in the sensor unit 106.
  • On-site physical calibration is where an operator visits and manually adjusts sensors in the sensor unit itself. Further it allows for historic measurements to also be calibrated.
  • a sensor may also be taken down, sent away for recalibration, and then reinstalled in a geographical region which is very time consuming.
  • the present technology uses mobile reference sensor units to improve efficiency and reduce cost.
  • the remote server 102 compensates for sensor drift in the raw sensor data stream by adjusting the measurements based on a first drift prediction.
  • the first drift prediction is generated by the remote server 102 based on a change in error between the first set of calibration measurements and the measurement values in the raw sensor data stream over the first calibration period.
  • the first drift prediction is generated by the remote server 102 through one or more of: extrapolation, linear regression, Kalman filter.
  • FIGs. 3A & 3B are graphs illustrating the calibration of a raw sensor data stream.
  • the inventors have recognized that the level of sensor drift in the one or more sensors of the sensor unit 106 varies with time, especially for internet of things sensors such as sensors on streetlamps in smart cities or at other locations. Drift occurs for a variety of reasons such as aging, temperature fluctuation, humidity fluctuation and others. Therefore, at some time after the first calibration period the level of sensor drift will change, and the adjustments applied by the remote server 102 using the first drift prediction will then introduce further error.
  • FIG. 3A is a graph 300 comprising value and time instant axes.
  • the “true” value is a high quality estimate of a ground truth value of the physical quantity and is shown in the FIGs. to aid understanding of the technology. In practice, when the technology is worked, the “true” value of the physical quantity may be unknown.
  • the values, timings and trend lines shown in FIGs. 3A-4B are for example and are not intended to be limiting.
  • the raw sensor data stream 306 shows a high level of sensor drift from the true value 310 around the first calibration period 302. This can be seen by the steeper gradient of the raw sensor data stream 306 compared with the true value which is essentially flat.
  • the adjustments performed based on the first set of calibration measurements will compensate for this high level of drift resulting in a reduction of the values over time for the first calibrated sensor data stream 308.
  • This provides a close fit between the first calibrated sensor data stream 308 and the true value 310 for several time instances even after the first calibration period 302.
  • the level of sensor drift reduces. With no further calibration measurements, the raw sensor data values are still adjusted for the first calibration stream as if the high level of sensor drift still applies. Therefore, as time progresses the first calibrated sensor data stream will have lower values than the true values of the physical quantity.
  • a second set of calibration measurements is received from the mobile reference sensor unit 104. This is the result of a revisit to the site of the sensor unit 106 by the mobile reference sensor unit 104 over a second calibration period that happens later than the first calibration period.
  • the revisit may be one of multiple periodic or irregular revisits based on scheduling of the mobile reference sensor unit 106. In one example the time between revisits is six months.
  • the measurement value represents the same physical quantity as that in the measurement values of the raw sensor data stream.
  • a second calibrated sensor data stream is generated this time using the second set of calibration measurements and the raw sensor data stream.
  • the measurement values in the stored raw sensor data stream are adjusted based on the error between the second set of calibration measurements and the raw sensor data stream values during the second calibration period.
  • the measurement times for the raw sensor data stream are preserved. This is performed in the same manner as during the generation of the first calibrated sensor data stream described above.
  • the remote server 102 To compensate for any new level of drift the remote server 102 generates a second drift prediction.
  • the second drift prediction is determined from the change in error between the second set of calibration measurements and the measurement values in the raw sensor data stream during the second calibration period. The second drift prediction is then used to adjust the values in the second calibrated sensor data stream.
  • the remote server identifies a time 406 between the first and second calibration times where there is a change in the sensor drift in the raw sensor data stream.
  • the identified time may be chosen as a time halfway or another designated point between the first and second calibration periods 302, 312.
  • an approach shown in FIG. 4A and method 500 can be used to identify the time as discussed in more detail below.
  • the remote server 102 generates a recalibrated sensor data stream. There are then two calibrated data streams stored at the remote server 102 both being based on the raw sensor data stream. The first calibrated sensor data stream comprising values adjusted based on the first set of calibration measurements and the second calibrated sensor data stream comprising values adjusted based on the second set of calibration measurements. The remote server 102 concatenates the calibrated data streams to generate a recalibrated data stream that better compensates for change in sensor drift level than either calibrated stream individually.
  • FIG. 3B shows one possible concatenation of the first and second calibrated data streams 308, 314 into the recalibrated sensor data stream.
  • the values of the first calibrated sensor data stream 308 are used up to the second calibration period.
  • the values of the second calibrated sensor data stream 314 are used for the remaining time instances.
  • the section of the recalibrated data stream during and after the second calibration period fits better to the true value than only using the first calibrated sensor data stream 308 from FIG. 3A.
  • the values of the first calibrated sensor data stream 308 do not fit the true value 310 because the change in drift occurs at a time between the first and second calibration periods 302, 312.
  • the time between visits from the mobile reference sensor 104 can be months or years this leaves a large period where the results are inaccurate.
  • the remote server 102 therefore uses the identified time as the point when the sensor drift changes. Up to the identified time the values from the first calibrated sensor data stream are used in the recalibrated sensor data stream. In the remaining time instances the values of the second sensor data stream are used. This is shown in FIG. 4B where it is illustrated that up to the identified time 406 the first calibrated data stream 308 is plotted and to the right of the identified time 406 the second calibrated sensor data stream 314 is plotted. This results in a more accurate sensor data stream because the measurement values are adjusted to compensate for the changed level of drift closer in time to where the change began, rather than waiting for the second calibration period 312.
  • FIG. 5 shows an example method 500 for identifying a time between the first and second calibration times where the level of sensor drift changes.
  • a first drift prediction is generated by the remote server 102 from the first set of calibration measurements.
  • the first drift prediction can be the same first drift prediction used to adjust the raw sensor data stream values into the first calibrated sensor data stream.
  • the first drift prediction is a predicted representation of the measurement values in the raw sensor data stream over time.
  • the remote server 102 generates the first drift prediction using the measurement values of the raw sensor data stream during the first calibration period and the first set of calibration measurements.
  • a second drift prediction is generated by the remote server 102 from the second set of calibration measurements. This can be the same second drift prediction used to adjust the raw sensor data stream values into the second calibrated sensor data stream.
  • the second drift prediction is a predicted representation of the “drifted” measurement values in the raw sensor data stream over time. The second drift prediction is generated using the measurement values of the raw sensor data stream during the second calibration period and the second set of calibration measurements.
  • Both the first and second drift predications are generated by the remote server 102 using one of linear regression, extrapolation, a trained machine learning model, or another artificial intelligence model.
  • the machine learning model is of any suitable type such as a random decision forest or neural network.
  • the machine learning model may be trained using supervised learning and labelled training data.
  • the labelled training data is pairs of raw sensor data and corresponding drift values obtained empirically.
  • the remote server module 102 identifies a point in time when the second drift prediction fits better to the measurement values of the raw sensor data module compared to the first drift prediction.
  • the remote server module 102 performs the identification using cross-correlation between the measurement values of the raw sensor data stream and the first and second drift predictions, or through other known techniques such as using a rule to check the magnitude of differences between the predicted values and measured values.
  • the remote server module 102 performs the identification of the point in time by computing an intersection of prediction lines 402 and 404 based on the equations of the two lines.
  • the first and second drift predictions of method 500 are illustrated graphically in FIG. 4A where graph 300 plots sensor measurement values against time.
  • the graph 300 further comprises the raw sensor data stream 306, the first drift prediction 402, the second drift prediction 404.
  • the first drift prediction 402 fits closely to the raw sensor data stream 306 until a change in sensor drift results in a decrease in overall gradient of the measurement values in the raw sensor data stream 306.
  • the second drift prediction 404 fits better to the raw sensor data stream 306.
  • the remote server 102 determines that the first time where the second drift prediction 404 fits better to the raw sensor data stream 306 is the identified time 406. Consequently, when generating the recalibrated sensor data stream the values before the identified time 406 are values from the first calibrated sensor data stream 308, and the values after are from the second calibrated sensor data stream 314.
  • a further revisit by the mobile reference sensor unit 104 is performed where the methods of FIG. 2 and 5 are repeated to regenerate the recalibrated sensor data stream. These repeated visits can continue throughout the operating life of the sensor unit 106 to reduce the error from future changes in sensor drift.
  • the second calibration measurements are received from a second mobile reference sensor unit. This allows for more flexible scheduling of mobile reference sensor units which may be attached to vehicles completing other tasks such as delivery or calibrating other sensor units.
  • FIG. 6 illustrates various components of an example remote server 102 in which embodiments of FIGs. 1-5 are implemented in some examples.
  • the remote server is a computing device of any suitable form such as a smart phone, a desktop computer, a tablet computer, or a laptop computer.
  • the remote server 102 comprises one or more processors 614 which are microprocessors, controllers, or any other suitable type of processors for processing computer executable instructions to control the operation of the device to perform the methods of FIGs. 2 and 5.
  • processors 614 include one or more fixed function blocks (also referred to as accelerators) which implement a part of the method of FIGs. 2 and 5 in hardware (rather than software or firmware). That is, the methods described herein are implemented in any one or more of software, firmware, hardware.
  • the remote server 102 has a data store holding an operating system 604, a raw sensor data stream 606, calibrated sensor data streams 608, the recalibrated sensor data stream 610 and optionally machine learning or other artificial intelligence models.
  • the remote server 102 has a communications interface 616 that allows connection to the sensor unit 106 and mobile reference sensor unit 104.
  • the communications interface is configurable to receive the raw sensor data stream and the first and second calibration measurements.
  • Platform software comprising an operating system 604 or any other suitable platform software is provided at the computing-based device to enable the methods of FIGs. 2 and 5 to be executed on the device.
  • the computer storage media data store 602 is shown within the computing-based device 102 it will be appreciated that the storage is, in some examples, distributed or located remotely and accessed via a network or other communication link (e.g. using communications interface 616).
  • the remote server module 102 also comprises an input/output controller 618 arranged to output display information to a display device 620 which may be separate from or integral to the remote server 102.
  • the display information may provide a graphical user interface.
  • the input/output controller 618 is also arranged to receive and process input from one or more devices, such as a user input device 620 (e.g. a mouse, keyboard, camera, microphone or other sensor).
  • a user input device 620 e.g. a mouse, keyboard, camera, microphone or other sensor.
  • the user input device 620 detects voice input, user gestures or other user actions.
  • the display device 622 also acts as the user input device 620 if it is a touch sensitive display device.
  • the input/output controller 618 outputs data to devices other than the display device in some examples.
  • any reference to 'an' item refers to one or more of those items.
  • the term 'comprising' is used herein to mean including the method blocks or elements identified, but that such blocks or elements do not comprise an exclusive list and an apparatus may contain additional blocks or elements and a method may contain additional operations or elements. Furthermore, the blocks, elements and operations are themselves not impliedly closed.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Arrangements For Transmission Of Measured Signals (AREA)

Abstract

A raw sensor data stream received from a sensor unit is stored. A first set of calibration measurements is received from a mobile reference sensor unit over a first calibration period. A calibrated sensor data stream is generated from the raw sensor data stream using the first set of calibration measurements. A second set of calibration measurements is received from the mobile reference sensor unit over a second calibration period later than the first calibration period. A second calibrated sensor data stream if generated from the raw sensor data stream using the second set of calibration measurements. A time is identified where there was a change of drift in the sensor unit. A recalibrated sensor data stream is generated comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.

Description

RECALIBRATING SENSOR VALUES
[0001] The present disclosure relates to recalibrating sensor values using a mobile reference sensor unit.
BACKGROUND
[0002] Electronic sensors are widely used to produce data representative of a physical phenomenon such as temperature, humidity etc. Generally, they produce an output that has a derived relationship with the physical phenomenon that can be used to convert an output signal of the electronic sensor into data. The output is the product of the interaction between an electronic circuit in the sensor and its surroundings. Certain components in the electronic circuits are susceptible to change their output when factors in their surroundings change. For example, amplifiers can be prone to output drift over time particularly when there is a variation in temperature. Their inclusion in a sensor means that, inevitably, the output will drift away from the derived relationship with the physical phenomena, leading to errors.
[0003] Sensors with high quality components such as low-drift amplifiers are expensive and do not eliminate drift completely. Sensor drift can be compensated for by calibrating the sensor using a separate control sensor that is known to be correct. Traditionally sensors could be adjusted following calibration by physically changing the circuit to compensate for the drift. One example is by adjusting a variable resistor until the output is changed closer to the expected value. This type of manual adjustment is cumbersome, time consuming, and expensive. Further, it cannot be applied retroactively to measurements already taken.
[0004] The amount of sensor drift varies in time with the physical condition of the sensor. Inevitably this means that no single calibration and adjustment will be sufficient for extended periods of use. In Internet of Things systems there can be thousands of sensors reporting at once. In smart city projects this is compounded by the large distances between the sensors with some sensors being required in places with low accessibility from the ground. With each sensor in these systems sensor drift occurs as described above. Calibrating these sensors manually with physical adjustments is extremely expensive and time consuming.
[0005] The examples described herein are not limited to examples which solve problems mentioned in this background section.
SUMMARY
[0006] Examples of preferred aspects and embodiments of the invention are as set out in the accompanying independent and dependent claims.
[0007] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.
[0008] According to a first aspect there is a computer-implemented method of recalibrating sensor values using a mobile reference sensor unit comprising: storing a raw sensor data stream received from a sensor unit; receiving a first set of calibration measurements from the mobile reference sensor unit over a first calibration period; generating a calibrated sensor data stream from the raw sensor data stream using the first set of calibration measurements; receiving a second set of calibration measurements from the mobile reference sensor unit over a second calibration period later than the first calibration period; generating a second calibrated sensor data stream from the raw sensor data stream using the second set of calibration measurements; identifying a time between the first and second calibration periods where there was a change of drift in the sensor unit using the raw sensor data stream and the first and second sets of calibration measurements; and generating a recalibrated sensor data stream comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.
In this way an efficient, scalable way of improving accuracy of sensors is achieved. The method scales to deployments of thousands or tens or thousands of sensors units such as in smart cities where the sensor units are on streetlamps or in other scenarios. Efficiency is gained by using a mobile reference sensor unit so that it is possible to have fewer mobile reference sensor units than sensor units. The mobile reference sensor units are able to move about to different ones of the sensor units giving scalability. Since the mobile reference sensor units may have more expensive or more sophisticated sensors than the sensor units, efficiency is gained (as there are fewer mobile reference sensor units). Also the sensor units themselves may be more vulnerable than the mobile reference sensor units.
[0009] Preferably, the sensor unit and the mobile sensor are both one of: air temperature, air quality, relative humidity, wind speed and road surface temperature sensors. Thus a wide range of sensor types is facilitated.
[0010] Preferably the sensor unit is at a fixed location. Where the sensor unit is at a fixed location there is a particular problem of scalability, since as the number of sensor units increases so does the geographical area of the site where the sensor units are deployed.
[0011] Preferably the sensor unit is periodically relocated and optionally generating the recalibrated sensor data stream is carried out only up to a time when the sensor unit was most recently relocated. Being able to periodically relocate the sensor unit brings flexibility, such as where a sensor on a portable street lamp is used. The process is able to operate well even where the sensor unit is periodically relocated. If a sensor unit is relocated, say, from an area where the air quality is good (e.g. low levels of particulates) to an area where the air quality is poor, there is likely to be a break in a time series of the sensor data stream (during the period of the relocation) followed by a significant jump in the values of the sensor data stream. In this case any recalibration is applied only back to the date/time the sensor unit started to resend data from its new location.
[0012] Preferably the mobile reference sensor unit is attached to a vehicle which is any of: an unmanned vehicle or an unmanned aerial UAV. By attaching the mobile reference sensor unit to a vehicle there is increased scalability since there may be many more sensor units than mobile reference sensors.
[0013] Preferably at least one of the first and second set of calibration measurements is collected within a threshold distance of the sensor unit. Improve accuracy may be achieved as a result since the calibration measurements generally have less noise where the sensor unit and mobile reference sensors are within a threshold distance of one another.
[0014] Preferably the mobile reference sensor unit provides more accurate values than the sensor unit. Where the mobile reference sensor unit is more accurate its measurements may be used for calibration with good results.
[0015] Preferably the mobile reference sensor unit is calibrated more frequently than the sensor unit. By calibrating the mobile reference sensor unit more frequently, it is possible to have better accuracy at the mobile reference sensor unit than at the sensor unit.
[0016] Preferably identifying the time further comprises: generating a first drift prediction for the raw sensor data stream using the first set of calibration measurements; generating a second drift prediction for the raw sensor data stream using the second set of calibration measurements; and identifying the earliest time between the first and second calibration periods when the second drift prediction fits better to the raw sensor data stream than the first drift prediction. This is an efficient and effective way to compute the identified time.
[0017] Preferably the first and second drift predictions are generated using any one or more of: linear regression, extrapolation, using a machine learning model, using another artificial intelligence model. In this way the drift predictions are computed quickly and accurately in a way which is scalable.
[0018] Preferably the method is performed by a remote server, and wherein the first and second calibration measurements are received through any of: a wireless communication, a wireless communication with higher quality than a communication link between the sensor unit and the remote server. By using a higher quality link for sending the calibration measurements than for sending the data from the sensor unit, it is possible to appropriately adjust the data from the sensor unit in an effective way.
[0019] Preferably the method comprises receiving any of the first and second calibration measurements from a store at the mobile reference sensor unit, in response to a communication link with the mobile reference sensor unit becoming available. In this way, the method is robust since calibration measurements are unlikely to be dropped in transmission between the mobile reference sensor unit and the remote server.
[0020] Preferably the communication link with the mobile reference sensor unit becomes available as a result of a wireless communication link becoming available or as a result of a fixed network connection becoming available in response to the mobile reference sensor unit docking into a unit with a fixed network connection. In this way the calibration data is effectively and efficiently sent to the remote server module.
[0021] In another aspect there is a computing device comprising: one or more processor; and a memory comprising instructions that when executed by the one or more processors, cause the processor to: store a raw sensor data stream received from a sensor in the memory; receive a first set of calibration measurements from the mobile reference sensor unit at a first calibration time; generate a calibrated sensor data stream from the raw sensor data stream using the first set of calibration measurements; receive a second set of calibration measurements from the mobile reference sensor unit at a second calibration time later than the first calibration time; generate a second calibrated sensor data stream from the raw sensor data stream using the second set of calibration measurements; identify a time between the first and second calibration times where there was a change of drift in the sensor unit using the raw sensor data stream and the first and second sets of calibration measurements; and generate a recalibrated sensor data stream comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.
[0022] In another aspect there is a system comprising: a sensor unit; a mobile reference sensor unit; and a computing device as described above.
[0023] It will also be apparent to anyone of ordinary skill in the art, that some of the preferred features indicated above as preferable in the context of one of the aspects of the disclosed technology indicated may replace one or more preferred features of other ones of the preferred aspects of the disclosed technology. Such apparent combinations are not explicitly listed above under each such possible additional aspect for the sake of conciseness.
[0024] Other examples will become apparent from the following detailed description, which, when taken in conjunction with the drawings, illustrate by way of example the principles of the disclosed technology.
BRIEF DESCRIPTION OF THE DRAWINGS
[0025] FIGs. 1A & 1 B are schematic diagrams illustrating a mobile reference sensor unit and a sensor unit communicating sensor measurements to a remote server through a communication network;
[0026] FIG. 2 is a flow diagram illustrating steps for recalibrating a sensor data stream using a mobile reference sensor;
[0027] FIGs. 3A & 3B are graphs illustrating the calibration of a raw sensor data stream;
[0028] FIG. 4A is a graph illustrating an example identification of a time where sensor drift has changed;
[0029] FIG. 4B is a graph illustrating a recalibrated sensor data stream;
[0030] FIG. 5 is a flow diagram illustrating part of FIG. 2 in greater detail;
[0031] FIG. 6 is a schematic diagram of a remote server module.
[0032] The accompanying drawings illustrate various examples. The skilled person will appreciate that the illustrated element boundaries (e.g., boxes, groups of boxes, or other shapes) in the drawings represent one example of the boundaries. It may be that in some examples, one element may be designed as multiple elements or that multiple elements may be designed as one element. Common reference numerals are used throughout the FIGs., where appropriate, to indicate similar features.
DETAILED DESCRIPTION
[0033] The following description is made for the purpose of illustrating the general principles of the present technology and is not meant to limit the inventive concepts claimed herein. As will be apparent to anyone of ordinary skill in the art, one or more or all of the particular features described herein in the context of one embodiment are also present in some other embodiment(s) and/or can be used in combination with other described features in various possible combinations and permutations in some other embodiment(s).
[0034] FIGs. 1 A and 1 B of the accompanying drawings show a system 100 comprising a remote server 102, a mobile reference sensor unit 104 and a sensor unit 106. The mobile reference sensor unit 104 and sensor unit 106 are both communicatively connected to the remote server 102 by way of a communications network 110 such as the internet, an intranet or any other communications network . The sensor unit 106 includes one or more sensors suitable to take measurements from an area surrounding the sensor unit 106. The sensor unit 106 also includes network equipment suitable for packetizing and sending the measurements to the remote server 102. The sensor unit 106 repeatedly measures at least one physical quantity from the surroundings and sends the measurements to the remote server 102 through the communications network 110. In some examples, the sensors included in the sensor unit 106 are one or more of: air temperature sensor, air quality sensor, humidity sensor, wind speed sensor, road surface temperature sensor. The remote server is a web server, a compute node of a communications network, or any other computing entity able to receive measurement data from the mobile reference sensor unit 104 and the sensor unit 106.
[0035] The sensor unit 106 is positioned at a desired site where it is desirable for the one or more sensors of the sensor unit 106 to measure the physical quantities of the surroundings. The sensor unit 106 is shown fixed to a streetlight in the example of FIG. 1A which shows a system 100 for recalibrating sensor data of the sensor unit 106. FIG. 1A is an example and is not intended to be limiting as the sensor unit 106 may be located at any location that is accessible by the mobile reference sensor unit 104. In some examples, the sensor unit 106 is moveable allowing periodic relocation to different measurement sites. Having a sensor unit 106 that can be relocated increases the efficiency of a smart city or other multi-site sensor system because it is possible to re-use the sensor unit 106 for different sites at different times. Measurements are taken by the sensor unit 106 periodically. In some examples that means measurements are taken at regular time intervals measured in seconds, minutes, hours etc. In an alternative example the measurements are taken at irregular time intervals. Irregular time intervals are preferable where the physical quantity is expected to vary more during certain periods than others, factors affecting this include night/day, seasonal variations, traffic levels etc.
[0036] FIG. 1A and FIG. 1B each show a single sensor unit 106. However, in practice there may be thousands or tens of thousands of sensor units such as in a smart city, or wider metropolitan area or across a county.
[0037] In various examples, measurements from the sensor unit 106 are received by the remote server 102 as a raw sensor data stream and are optionally stored at the remote server 102 or another location. The raw sensor data stream includes for each measurement a measurement value and measurement time. The raw sensor data stream is stored in time order; that is, the sensor data stream is a time series. In some examples the measurement value is the output from the one or more sensors for example a voltage or current value. In an alternative embodiment the measurement value is converted at the sensor unit 106 into data expressing the physical quantity, for example degrees Celsius for air or road temperature.
[0038] The inventors have recognized that, over time errors increase in the raw sensor data stream because of sensor drift in the one or more sensors in the sensor unit 106. As discussed above, this is an inevitable consequence of the components found in the electric circuits of the one or more sensors. The time variant nature of sensor drift means that the error size between the true value and the measurements will change over time. Consequently, the stored raw sensor data stream at the remote server 102 has a variable error across time. To address this the inventors have developed a way of using a mobile reference sensor unit 104 to collect measurements separately from the sensor unit 106. The separately collected measurements from the mobile reference sensor unit 104 are used for calibrating and recalibrating the raw sensor data stream of the sensor unit 106. Since the measurements from the mobile reference sensor unit 104 are independent of the measurements of the raw sensor data stream it is possible to achieve accurate calibration and recalibration. The term “calibration” is used to refer to correcting error in sensor data measurements. The term “recalibration” is used to refer to correcting error in sensor data measurements which have already been calibrated at least once.
[0039] The mobile reference sensor unit 104 comprises one or more sensors. The one or more sensors comprise at least one sensor that measures a same physical quantity as the sensor unit 106. The mobile reference sensor unit 104 comprises sensors that are expected to produce smaller errors than those in the sensor unit 106. In one example this is achieved by using higher quality sensors in the mobile reference sensor unit 106 than in the sensor unit 106. Alternatively or additionally this is achieved by more frequently calibrating the one or more sensors of the mobile reference sensor unit 104. The mobile reference sensor unit 104 also comprises network equipment capable of sending calibration measurements to the remote server 102 through the communications network 110. In some cases the mobile reference sensor unit 104 has a memory to store the calibration measurements. When the mobile reference sensor unit 104 moves to a location where communication with the remote server 102 is possible, the calibration measurements are sent to the remote server 102.
[0040] The mobile reference sensor unit 104 is attached to a vehicle or other movable entity such as a temporary tower. In one example shown in FIG. 1A the mobile reference sensor unit 104 is attached to a wheeled vehicle which can be manned or unmanned. Alternatively, as shown in FIG. 1 B the mobile reference sensor unit 104 is an aerial vehicle, such as an unmanned aerial vehicle (UAV). By attaching the mobile reference sensor 104 to a vehicle the system 100 has improved efficiency because the mobile reference sensor unit 104 may provide calibration measurements for multiple sensors. The use of unmanned vehicles reduces operator burden for the system 100 because a human operator is not required to calibrate the sensor unit 106. Further, the use of a UAV produces a more efficient system 100 because it allows access to the site where the sensor unit 106 is located even when the site is inaccessible from the ground. [0041] To collect the calibration measurements, the mobile reference sensor unit 104 moves to visit the site of the sensor unit 106. The mobile reference sensor unit 104 moves to the sensor unit 106 in any suitable way, such as by moving to a known location following instructions from the remote server 102. In some cases the sensor unit 106 emits a beacon signal and when the mobile reference sensor unit 104 detects the beacon signal it moves towards the sensor unit 106.
[0042] Once positioned near to the sensor unit 106, the mobile reference sensor unit 104 begins to measure at least one of the same physical quantities as measured by the sensor unit 106. In some examples the mobile reference sensor unit 104 is moved within a threshold distance of the sensor unit 106 based on the physical quantity being measured. The threshold distance is around 1 to 2 metres in some cases. In an example the mobile reference sensor unit 104 moves within two metres of the sensor unit 106 before beginning to take measurements. The mobile reference sensor unit 104 measures the physical quantity over the course of a calibration period. The calibration period being set to a length such that it allows the mobile reference sensor unit 104 to record a plurality of calibration measurements that correspond to measurements taken by the sensor unit 106.
[0043] The remote server 102 receives both the raw sensor data stream from the sensor unit 106 and the plurality of calibration measurements from the mobile reference sensor unit 104. In some embodiments both the raw sensor data stream and the plurality of calibration measurements are received through a wireless link via communication network 110. In some examples the plurality of calibration measurements are received through a wireless link of higher quality than that used to send the raw sensor data stream to the remote server 102. In some examples the mobile reference sensor unit 104 stores the calibration measurements until a connection to the communications network 110 with the remote server 102 becomes available. In some examples the mobile reference sensor unit 104 is configured to take the measurements whilst physically proximate to the sensor unit 106, store the measurements in it’s memory, move to another location and dock into a unit that has a fixed wired connection to the communications network 110. The fixed wired connection may be of higher quality than the communications link between the sensor unit 106 and the remote server 102. In this way the fixed sensor unit 102 may be low cost such as with lower quality data transmission equipment.
[0044] The remote server 102 comprises a memory configured to store the received raw sensor data stream and calibration measurements. The remote server 102 uses a first set of calibration measurements to calibrate the raw sensor data stream stored from the sensor unit 106. As discussed in more detail below this first calibration eventually introduces further error by attempting to compensate for a level of drift that has since changed. Recalibration is therefore done. To carry out recalibration, the visit of the mobile reference sensor unit 104 to the site of sensor unit 106, shown in FIGs. 1A and 1 B, is repeated. Having received a second set of calibration measurements the remote server 102 recalibrates the raw sensor data stream. This recalibration is described in more detail regarding FIG. 2 below.
[0045] FIG. 2 illustrates a method 200 for recalibrating sensor values using a mobile reference sensor unit 104 such as the one described above in system 100. The method is performed at a remote server such as the remote server 102.
[0046] At step 202 a raw sensor data stream is stored having been received from a sensor unit such as the sensor unit 106. Each measurement in the raw sensor data stream includes a measurement value and measurement time. The time between each measurement value is regular or irregularly spaced. The measurement values represent values of a physical quantity measured by one or more of the sensors in the sensor unit 106.
[0047] At step 204 a first set of calibration measurements is received. The first set of calibration measurements having been collected during a first visit of the mobile reference sensor unit 104 to the site of the sensor unit 106. The first set of calibration measurements is sent from the sensor unit 106 to the remote server 102 over the communications network 110. Each measurement in the first set of calibration measurements includes a measurement value and a measurement time recorded by the mobile reference sensor unit 104. The measurement values in the first set of calibration measurements represent the same physical quantity as those of the raw sensor data stream.
[0048] At step 206 a first calibrated data stream is generated using the first set of calibration measurements. The first calibrated sensor data stream includes the measurement values of the raw sensor data stream adjusted by the remote server 102, based on the first set of calibration measurements. In generating the first calibrated sensor data stream, the amount of adjustment applied to the measurement values in the raw sensor data stream is based on an error between the first set of calibration measurements and corresponding values of the raw sensor data stream. The error is computed in any suitable way such as by computing an average difference between the corresponding values in a specified time window, computing a mean squared difference between the corresponding values, computing a median difference between the corresponding values, computing a mode of the differences between the corresponding values or in other ways. For example if the first set of calibration measurements are determined by the remote server 102 to be larger than the measurement values for the raw sensor data stream at the nearest times then an offset adjustment is made to increase the values for all raw sensor data stream values. Historic and subsequent measurement values for the raw sensor data stream received from the sensor unit 106, are then also adjusted and included in the first calibrated data stream. The measurement time values are not adjusted. [0049] Calibrating the raw sensor data stream to produce the first calibrated sensor data stream in this way is more efficient and less expensive than on-site physical calibration in the sensor unit 106. On-site physical calibration is where an operator visits and manually adjusts sensors in the sensor unit itself. Further it allows for historic measurements to also be calibrated. A sensor may also be taken down, sent away for recalibration, and then reinstalled in a geographical region which is very time consuming. The present technology uses mobile reference sensor units to improve efficiency and reduce cost.
[0050] When adjusting the measurement values, the remote server 102 compensates for sensor drift in the raw sensor data stream by adjusting the measurements based on a first drift prediction. The first drift prediction is generated by the remote server 102 based on a change in error between the first set of calibration measurements and the measurement values in the raw sensor data stream over the first calibration period. The first drift prediction is generated by the remote server 102 through one or more of: extrapolation, linear regression, Kalman filter.
[0051] FIGs. 3A & 3B are graphs illustrating the calibration of a raw sensor data stream. The inventors have recognized that the level of sensor drift in the one or more sensors of the sensor unit 106 varies with time, especially for internet of things sensors such as sensors on streetlamps in smart cities or at other locations. Drift occurs for a variety of reasons such as aging, temperature fluctuation, humidity fluctuation and others. Therefore, at some time after the first calibration period the level of sensor drift will change, and the adjustments applied by the remote server 102 using the first drift prediction will then introduce further error. The effect of this change of sensor drift on the first calibrated sensor data stream is shown in FIG. 3A which is a graph 300 comprising value and time instant axes. Shown on the graph are a first calibration period 302, a raw sensor data stream 306, a first calibrated sensor data stream 308 and a “true” value for a physical quantity 310. The “true” value is a high quality estimate of a ground truth value of the physical quantity and is shown in the FIGs. to aid understanding of the technology. In practice, when the technology is worked, the “true” value of the physical quantity may be unknown. The values, timings and trend lines shown in FIGs. 3A-4B are for example and are not intended to be limiting.
[0052] The raw sensor data stream 306 shows a high level of sensor drift from the true value 310 around the first calibration period 302. This can be seen by the steeper gradient of the raw sensor data stream 306 compared with the true value which is essentially flat. The adjustments performed based on the first set of calibration measurements will compensate for this high level of drift resulting in a reduction of the values over time for the first calibrated sensor data stream 308. This provides a close fit between the first calibrated sensor data stream 308 and the true value 310 for several time instances even after the first calibration period 302. [0053] However, after the first calibration period the level of sensor drift reduces. With no further calibration measurements, the raw sensor data values are still adjusted for the first calibration stream as if the high level of sensor drift still applies. Therefore, as time progresses the first calibrated sensor data stream will have lower values than the true values of the physical quantity.
[0054] At step 208, to address the erroneous calibration for changeable levels of drift, a second set of calibration measurements is received from the mobile reference sensor unit 104. This is the result of a revisit to the site of the sensor unit 106 by the mobile reference sensor unit 104 over a second calibration period that happens later than the first calibration period. The revisit may be one of multiple periodic or irregular revisits based on scheduling of the mobile reference sensor unit 106. In one example the time between revisits is six months. As with the first set of calibration measurements the measurement value represents the same physical quantity as that in the measurement values of the raw sensor data stream.
[0055] At step 210, a second calibrated sensor data stream is generated this time using the second set of calibration measurements and the raw sensor data stream. To generate the second calibrated sensor data stream, the measurement values in the stored raw sensor data stream are adjusted based on the error between the second set of calibration measurements and the raw sensor data stream values during the second calibration period. The measurement times for the raw sensor data stream are preserved. This is performed in the same manner as during the generation of the first calibrated sensor data stream described above.
[0056] To compensate for any new level of drift the remote server 102 generates a second drift prediction. The second drift prediction is determined from the change in error between the second set of calibration measurements and the measurement values in the raw sensor data stream during the second calibration period. The second drift prediction is then used to adjust the values in the second calibrated sensor data stream.
[0057] At step 212, the remote server identifies a time 406 between the first and second calibration times where there is a change in the sensor drift in the raw sensor data stream. The identified time may be chosen as a time halfway or another designated point between the first and second calibration periods 302, 312. Alternatively, an approach shown in FIG. 4A and method 500 can be used to identify the time as discussed in more detail below.
[0058] At step 214, the remote server 102 generates a recalibrated sensor data stream. There are then two calibrated data streams stored at the remote server 102 both being based on the raw sensor data stream. The first calibrated sensor data stream comprising values adjusted based on the first set of calibration measurements and the second calibrated sensor data stream comprising values adjusted based on the second set of calibration measurements. The remote server 102 concatenates the calibrated data streams to generate a recalibrated data stream that better compensates for change in sensor drift level than either calibrated stream individually.
[0059] FIG. 3B shows one possible concatenation of the first and second calibrated data streams 308, 314 into the recalibrated sensor data stream. In this concatenation the values of the first calibrated sensor data stream 308 are used up to the second calibration period. Then the values of the second calibrated sensor data stream 314 are used for the remaining time instances. As shown on the graph the section of the recalibrated data stream during and after the second calibration period fits better to the true value than only using the first calibrated sensor data stream 308 from FIG. 3A. However, for a period between the first and second calibration periods 302, 312 the values of the first calibrated sensor data stream 308 do not fit the true value 310 because the change in drift occurs at a time between the first and second calibration periods 302, 312. As the time between visits from the mobile reference sensor 104 can be months or years this leaves a large period where the results are inaccurate.
[0060] The remote server 102 therefore uses the identified time as the point when the sensor drift changes. Up to the identified time the values from the first calibrated sensor data stream are used in the recalibrated sensor data stream. In the remaining time instances the values of the second sensor data stream are used. This is shown in FIG. 4B where it is illustrated that up to the identified time 406 the first calibrated data stream 308 is plotted and to the right of the identified time 406 the second calibrated sensor data stream 314 is plotted. This results in a more accurate sensor data stream because the measurement values are adjusted to compensate for the changed level of drift closer in time to where the change began, rather than waiting for the second calibration period 312.
[0061] FIG. 5 shows an example method 500 for identifying a time between the first and second calibration times where the level of sensor drift changes. At step 502 a first drift prediction is generated by the remote server 102 from the first set of calibration measurements. The first drift prediction can be the same first drift prediction used to adjust the raw sensor data stream values into the first calibrated sensor data stream. The first drift prediction is a predicted representation of the measurement values in the raw sensor data stream over time. The remote server 102 generates the first drift prediction using the measurement values of the raw sensor data stream during the first calibration period and the first set of calibration measurements.
[0062] At step 504 a second drift prediction is generated by the remote server 102 from the second set of calibration measurements. This can be the same second drift prediction used to adjust the raw sensor data stream values into the second calibrated sensor data stream. The second drift prediction is a predicted representation of the “drifted” measurement values in the raw sensor data stream over time. The second drift prediction is generated using the measurement values of the raw sensor data stream during the second calibration period and the second set of calibration measurements.
[0063] Both the first and second drift predications are generated by the remote server 102 using one of linear regression, extrapolation, a trained machine learning model, or another artificial intelligence model.
[0064] In the case that a trained machine learning model is used, the machine learning model is of any suitable type such as a random decision forest or neural network. The machine learning model may be trained using supervised learning and labelled training data. In an example, the labelled training data is pairs of raw sensor data and corresponding drift values obtained empirically.
[0065] Other types of artificial intelligence model that may be used include but are not limited to: rule-based systems, Bayesian graphical models.
[0066] At step 506 the remote server module 102 identifies a point in time when the second drift prediction fits better to the measurement values of the raw sensor data module compared to the first drift prediction. The remote server module 102 performs the identification using cross-correlation between the measurement values of the raw sensor data stream and the first and second drift predictions, or through other known techniques such as using a rule to check the magnitude of differences between the predicted values and measured values. In some examples the remote server module 102 performs the identification of the point in time by computing an intersection of prediction lines 402 and 404 based on the equations of the two lines.
[0067] The first and second drift predictions of method 500 are illustrated graphically in FIG. 4A where graph 300 plots sensor measurement values against time. The graph 300 further comprises the raw sensor data stream 306, the first drift prediction 402, the second drift prediction 404. As shown in the graph 300 the first drift prediction 402 fits closely to the raw sensor data stream 306 until a change in sensor drift results in a decrease in overall gradient of the measurement values in the raw sensor data stream 306. After the change in sensor drift the second drift prediction 404 fits better to the raw sensor data stream 306. The remote server 102 determines that the first time where the second drift prediction 404 fits better to the raw sensor data stream 306 is the identified time 406. Consequently, when generating the recalibrated sensor data stream the values before the identified time 406 are values from the first calibrated sensor data stream 308, and the values after are from the second calibrated sensor data stream 314.
[0068] In some embodiments a further revisit by the mobile reference sensor unit 104 is performed where the methods of FIG. 2 and 5 are repeated to regenerate the recalibrated sensor data stream. These repeated visits can continue throughout the operating life of the sensor unit 106 to reduce the error from future changes in sensor drift. [0069] In some examples, the second calibration measurements are received from a second mobile reference sensor unit. This allows for more flexible scheduling of mobile reference sensor units which may be attached to vehicles completing other tasks such as delivery or calibrating other sensor units.
[0070] FIG. 6 illustrates various components of an example remote server 102 in which embodiments of FIGs. 1-5 are implemented in some examples. The remote server is a computing device of any suitable form such as a smart phone, a desktop computer, a tablet computer, or a laptop computer.
[0071] The remote server 102 comprises one or more processors 614 which are microprocessors, controllers, or any other suitable type of processors for processing computer executable instructions to control the operation of the device to perform the methods of FIGs. 2 and 5. In some examples, a system on a chip architecture is used, the processors 614 include one or more fixed function blocks (also referred to as accelerators) which implement a part of the method of FIGs. 2 and 5 in hardware (rather than software or firmware). That is, the methods described herein are implemented in any one or more of software, firmware, hardware. The remote server 102 has a data store holding an operating system 604, a raw sensor data stream 606, calibrated sensor data streams 608, the recalibrated sensor data stream 610 and optionally machine learning or other artificial intelligence models. The remote server 102 has a communications interface 616 that allows connection to the sensor unit 106 and mobile reference sensor unit 104. The communications interface is configurable to receive the raw sensor data stream and the first and second calibration measurements. Platform software comprising an operating system 604 or any other suitable platform software is provided at the computing-based device to enable the methods of FIGs. 2 and 5 to be executed on the device. Although the computer storage media (data store 602) is shown within the computing-based device 102 it will be appreciated that the storage is, in some examples, distributed or located remotely and accessed via a network or other communication link (e.g. using communications interface 616).
[0072] The remote server module 102 also comprises an input/output controller 618 arranged to output display information to a display device 620 which may be separate from or integral to the remote server 102. The display information may provide a graphical user interface. The input/output controller 618 is also arranged to receive and process input from one or more devices, such as a user input device 620 (e.g. a mouse, keyboard, camera, microphone or other sensor). In some examples the user input device 620 detects voice input, user gestures or other user actions. In an embodiment the display device 622 also acts as the user input device 620 if it is a touch sensitive display device. The input/output controller 618 outputs data to devices other than the display device in some examples. [0073] Any reference to 'an' item refers to one or more of those items. The term 'comprising' is used herein to mean including the method blocks or elements identified, but that such blocks or elements do not comprise an exclusive list and an apparatus may contain additional blocks or elements and a method may contain additional operations or elements. Furthermore, the blocks, elements and operations are themselves not impliedly closed.
[0074] The steps of the methods described herein may be carried out in any suitable order, or simultaneously where appropriate. The arrows between boxes in the FIGs. show one example sequence of method steps but are not intended to exclude other sequences or the performance of multiple steps in parallel. Additionally, individual blocks may be deleted from any of the methods without departing from the spirit and scope of the subject matter described herein. Aspects of any of the examples described above may be combined with aspects of any of the other examples described to form further examples without losing the effect sought. Where elements of the FIGs. are shown connected by arrows, it will be appreciated that these arrows show just one example flow of communications (including data and control messages) between elements. The flow between elements may be in either direction or in both directions.
[0075] Where the description has explicitly disclosed in isolation some individual features, any apparent combination of two or more such features is considered also to be disclosed, to the extent that such features or combinations are apparent and capable of being carried out based on the present specification as a whole in the light of the common general knowledge of a person skilled in the art, irrespective of whether such features or combinations of features solve any problems disclosed herein. In view of the foregoing description, it will be evident to a person skilled in the art that various modifications may be made within the scope of the invention.

Claims

1. A computer-implemented method of recalibrating sensor values using a mobile reference sensor unit comprising: storing a raw sensor data stream received from a sensor unit; receiving a first set of calibration measurements from the mobile reference sensor unit over a first calibration period; generating a calibrated sensor data stream from the raw sensor data stream using the first set of calibration measurements; receiving a second set of calibration measurements from the mobile reference sensor unit over a second calibration period later than the first calibration period; generating a second calibrated sensor data stream from the raw sensor data stream using the second set of calibration measurements; identifying a time between the first and second calibration periods where there was a change of drift in the sensor unit using the raw sensor data stream and the first and second sets of calibration measurements; and generating a recalibrated sensor data stream comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.
2. The method according to claim 1 wherein the sensor unit and the mobile sensor are both one of air temperature, air quality, relative humidity, wind speed and road surface temperature sensors.
3. The method according to either of claims 1 or 2 wherein the sensor unit is at a fixed location.
4. The method according to claims 1 or 2 wherein the sensor unit is periodically relocated and optionally wherein generating the recalibrated sensor data stream is carried out only up to a time when the sensor unit was most recently relocated.
5. The method according to any of the previous claims wherein the mobile reference sensor unit is attached to a vehicle which is any of: an unmanned vehicle or an unmanned aerial UAV.
6. The method according to any of the preceding claims wherein at least one of the first and second set of calibration measurements is collected within a threshold distance of the sensor unit.
7. The method according to any of the previous claims wherein the mobile reference sensor unit provides more accurate values than the sensor unit.
8. The method of claim 7 wherein the mobile reference sensor unit is calibrated more frequently than the sensor unit.
9. The method according to any of the previous claims wherein identifying the time further comprises: generating a first drift prediction for the raw sensor data stream using the first set of calibration measurements; generating a second drift prediction for the raw sensor data stream using the second set of calibration measurements; and identifying the earliest time between the first and second calibration periods when the second drift prediction fits better to the raw sensor data stream than the first drift prediction.
10. The method according to claim 9 wherein the first and second drift predictions are generated using any one or more of: linear regression, extrapolation, using a machine learning model, using another artificial intelligence model.
11. The method according to any of the previous claims performed by a remote server, and wherein the first and second calibration measurements are received through any of: a wireless communication, a wireless communication with higher quality than a communication link between the sensor unit and the remote server.
12. The method of claim 11 comprising receiving any of the first and second calibration measurements from a store at the mobile reference sensor unit, in response to a communication link with the mobile reference sensor unit becoming available.
13. The method according to claim 12 wherein the communication link with the mobile reference sensor unit becomes available as a result of a wireless communication link becoming available or as a result of a fixed network connection becoming available in response to the mobile reference sensor unit docking into a unit with a fixed network connection.
14. A computing device comprising: one or more processor; and a memory comprising instructions that when executed by the one or more processors, cause the processor to: store a raw sensor data stream received from a sensor in the memory; receive a first set of calibration measurements from the mobile reference sensor unit at a first calibration time; generate a calibrated sensor data stream from the raw sensor data stream using the first set of calibration measurements; receive a second set of calibration measurements from the mobile reference sensor unit at a second calibration time later than the first calibration time; generate a second calibrated sensor data stream from the raw sensor data stream using the second set of calibration measurements; identify a time between the first and second calibration times where there was a change of drift in the sensor unit using the raw sensor data stream and the first and second sets of calibration measurements; and generate a recalibrated sensor data stream comprising: values from the first calibrated sensor data stream up to the identified time; and values from the second calibrated sensor data stream after the identified time.
15. A system comprising: a sensor unit; a mobile reference sensor unit; and a computing device according to claim 14.
EP23818020.2A 2022-12-22 2023-12-04 Recalibrating sensor values Pending EP4639097A1 (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
GBGB2219548.1A GB202219548D0 (en) 2022-12-22 2022-12-22 recalibrating sensor values
EP22216130 2022-12-22
PCT/EP2023/084160 WO2024132478A1 (en) 2022-12-22 2023-12-04 Recalibrating sensor values

Publications (1)

Publication Number Publication Date
EP4639097A1 true EP4639097A1 (en) 2025-10-29

Family

ID=89119437

Family Applications (1)

Application Number Title Priority Date Filing Date
EP23818020.2A Pending EP4639097A1 (en) 2022-12-22 2023-12-04 Recalibrating sensor values

Country Status (2)

Country Link
EP (1) EP4639097A1 (en)
WO (1) WO2024132478A1 (en)

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2457660A (en) * 2008-02-19 2009-08-26 Sphere Medical Ltd Methods of calibrating a sensor in a patient monitoring system
GB2512842A (en) * 2013-04-08 2014-10-15 Sphere Medical Ltd Sensor calibration method and apparatus
US11781890B2 (en) * 2017-12-29 2023-10-10 Intel Corporation Method, a circuit and a system for environmental sensing
FI128841B (en) * 2018-03-22 2021-01-15 Univ Helsinki Sensor calibration

Also Published As

Publication number Publication date
WO2024132478A1 (en) 2024-06-27

Similar Documents

Publication Publication Date Title
US11494618B2 (en) Anomaly detection using deep learning on time series data
US12056936B2 (en) Model generating apparatus, method and program, and prediction apparatus
JP7003953B2 (en) Learning equipment, estimation equipment, data generation equipment, learning methods, and learning programs
EP3769047B1 (en) Sensor calibration
US12487569B2 (en) Method of performing a process and optimizing control signals used in the process
US10936940B2 (en) Arrangement and method for predicting road friction within a road network
US11715369B2 (en) Latent space model for road networks to predict time-varying traffic
CN110617916B (en) A method and device for calibrating an air pressure sensor
US11693434B2 (en) Water quality monitoring system and method thereof
US20230074474A1 (en) Parameter adjustment apparatus, inference apparatus, parameter adjustment method, and computer-readable storage medium storing a parameter adjustment program
CN108509970A (en) Learning method, corresponding system, equipment and computer program product
US11475310B1 (en) Training network to minimize worst-case error
US20180144266A1 (en) Learning apparatus and method for learning a model corresponding to real number time-series input data
US20180232637A1 (en) Machine learning method for optimizing sensor quantization boundaries
CN115130658B (en) Reservoir landslide displacement prediction method and device based on nonlinear combination model
CN119191010A (en) Method and device for self-correcting air pressure height of elevator based on mechanical state
WO2024132478A1 (en) Recalibrating sensor values
CN113821535A (en) A method and device for model training, computer equipment and storage medium
CN117273217A (en) A space debris orbit prediction method, system and electronic equipment
WO2021042401A1 (en) Method and device for traffic light control
CN120141396A (en) A method and system for monitoring transmission tower settlement based on Beidou positioning
KR102238472B1 (en) Method for correcting error and sensor system
CN114503210A (en) Monitoring performance of predictive computer-implemented models
CN116881768A (en) Data processing method, device, computer equipment and storage medium
Toukourou et al. Flash flood forecasting by statistical learning in the absence of rainfall forecast: a case study

Legal Events

Date Code Title Description
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: UNKNOWN

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE

PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20250620

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC ME MK MT NL NO PL PT RO RS SE SI SK SM TR

P01 Opt-out of the competence of the unified patent court (upc) registered

Free format text: CASE NUMBER: UPC_APP_0003675_4639097/2026

Effective date: 20260202

DAV Request for validation of the european patent (deleted)
DAX Request for extension of the european patent (deleted)