An NTC temperature sensor calibration method based on edge computing

Abstract

The application discloses an NTC temperature sensor calibration method based on edge computing, which deploys an edge computing network on a calibration site, and a device end sends data of a device to be calibrated and a standard device to an edge node according to preset rules through a TCP / IP protocol; the edge node analyzes the reported data and temporarily stores the data in a local database; the edge node selects optimal calibration feature points through a genetic algorithm; according to the feature points, the device to be calibrated and the standard device are calibrated at the feature points, and the measurement temperature of the feature points of the temperature sensor is accurately calculated through the calibration data; the edge node sends the calibration data and the calibration result to a cloud end to realize cloud storage of the calibration record, and sends a characteristic function to the device end to complete calibration. Compared with a traditional calibration mode, the calibration period can be greatly shortened, the bandwidth consumption of a cloud server is effectively saved, the calculation pressure of the cloud end is reduced, and the precision and accuracy of the NTC temperature sensor can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of temperature calibration and measurement, and particularly relates to a method for calibrating an NTC temperature sensor by using edge computing technology. BACKGROUND

[0002] A traditional temperature calibration method, i.e., laboratory calibration, is that a customer periodically sends his own temperature meter to be calibrated to a higher-level calibration institution, calibration professionals measure and compare the temperature meter with a high-level standard platinum resistance in a controllable calibration laboratory, and the calibrated temperature meter is sent back to the relevant customer unit, and a calibration certificate is signed and sent to the relevant customer unit. This traditional method has the disadvantages of long calibration period, high uncertainty, great management difficulty, and low efficiency, which finally affects the normal production progress of modern industry.

[0003] In view of the above problems, the present application uses edge computing technology to complete data fitting with a standard device at the edge of the device and establish an error compensation model, so as to reduce the error of the temperature sensor when measuring temperature at the software level. SUMMARY

[0004] The present application aims at the disadvantages of long calibration period, great management difficulty, and low efficiency of the traditional temperature sensor calibration method, and proposes an NTC temperature sensor calibration method based on edge computing.

[0005] The purpose of the present application is achieved by the following technical scheme: an NTC temperature sensor calibration method based on edge computing, which comprises the following steps: step 1: deploying an edge computing network at a calibration site;

[0006] Step 2: after starting the calibration process, the device end sends the temperature data of the NTC temperature sensor to be calibrated and the standard device containing the detection index to the edge node through the TCP / IP protocol;

[0007] Step 3: the edge node analyzes the received data

[0008] Step 4: the edge node selects the best feature point for calibration by using a genetic algorithm;

[0009] Step 5: according to the feature point obtained in step 4, the NTC temperature sensor to be calibrated and the standard device are calibrated at the feature point, and the measurement temperature of the temperature sensor feature point is accurately calculated through the calibration data;

[0010] Step 6: the edge node sends the calibration data and the calibration result to the cloud end by using the MQTT protocol to realize cloud storage of the calibration record, and sends the characteristic function to the device end by using the TCP / IP protocol to complete calibration.

[0011] Further, the detection index contained in step 2 is specifically resistance, temperature, timestamp, device number and ambient temperature.

[0012] Further, in order to reduce the time cost spent on calibration in step 4, a genetic algorithm is used to select the optimal feature point for calibration, and the specific steps of obtaining the optimal feature point are as follows:

[0013] (4.1) Establishment of genes and selection of feature points: Abstract each temperature data sent by the device end as a gene, define x q qth test point is a feature point, when x q = 1 indicates that the qth test point is selected as a feature point, otherwise when x q = 0 indicates that the qth test point is not selected as a feature point.

[0014] (4.2) Definition of fitness function: Based on the given optimization target, the fitness function is used to measure the quality of chromosomes in the population; Since the goal of decision is to reduce the total time cost, the fitness function is selected as the square inverse of the total time cost C total of chromosomes,

[0015] (4.2.1) Single measurement time cost: The time cost required to measure a temperature value is denoted as C q

[0016] (4.2.2) Calibration error time cost:

[0017]

[0018] Symbol c j The time cost value corresponding to the single point error. Where i represents the sensor number, j represents the test point, T i, represents the actual temperature value, represents the sensor reading;

[0019] The calibration error time cost of a single sample body is calculated by

[0020] (4.2.3) Sample individual calibration time cost: The calibration time cost C i of a sample individual is the sum of the single measurement time cost and the error calibration time cost;

[0021] C i = C s +C q ·n i

[0022] n i represents the number of measurement points in the calibration process of the sample individual;​​

[0023] (4.2.4) Total cost: Total cost C total is the average of the labeled time cost for all individuals, and M is the total number of samples;

[0024]

[0025] (4.3) Crossover operation: Calculate offspring c1 and c2 from parents p1 and p2 using binary crossover with simulation; calculate the parameters of the offspring by selecting a random number u in [0, 1) The offspring can be represented as

[0026] (4.4) Mutation operation: For the simulation of the genetic process, if a mutation occurs in an individual, each feature of the individual is multiplied by a random number in the range [1-m rgnge , 1+m range ], m range is calculated independently for each feature as a multiplication factor, and m range ∈ [0.01, 0.3, 10% of the best individuals are selected from the current population by roulette selection to enter the next generation of iterations, so that the best individuals can be retained to the maximum extent;

[0027] (4.5) Termination condition: Genetic iteration is terminated when the following range is exceeded:

[0028]

[0029] σ 2 (r i ) is the variance of the i-th temperature feature point r i , σ 2 (W i ) is the variance of the i-th feature point combination W i , cov(r i -r j ) is the covariance of feature points i, j, is the minimum input temperature, is the maximum input temperature.

[0030] Further, in step 5, a third-order polynomial fitting formula of the Steinhart-Hart algorithm is selected to fit and analyze the temperature data, specifically as follows:

[0031] A third-order polynomial fitting formula is selected to fit the RT table data of the NTC thermistor: where T is the open temperature scale, unit K, R is the resistance, unit KΩ, A, B, C are constant coefficients; by analyzing the sum of squares of errors That is:

[0032]

[0033] In the formula, R i is the i-th resistance, T i is the R i matched temperature value; constant a i, Substitute and bring into the polynomial fitting formula to obtain the constant matrix:

[0034]

[0035] The constant coefficient of the fitting formula is:

[0036] A=(a23^2*b1-a12*a23*b3+a13*a22*b3-a13*a23*b2+a12*a33*b2-a22*a33*b1) / (a33*a12^2-2*a12*a13*a23+a22*a13^2+a11*a23^2-a11*a22*a33)

[0037] B=(a13^2*b2-a12*a13*b3+a11*a23*b3-a13*a23*b1-a11*a33*b2+a12*a33*b1) / (a33*a12^2-2*a12*a13*a23+a22*a13^2+a11*a23^2-a11*a22*a33)

[0038] C=(a12^2*b3-a12*a13*b2-a11*a22*b3+a11*a23*b2-a12*a23*b1+a13*a22*b1) / (a33*a12^2-2*a12*a13*a23+a22*a13^2+a11*a23^2-a11*a22*a33)

[0039] Thus, the temperature standard value at a certain resistance value is calculated.

[0040] Further, the edge gateway in step 5 calculates the temperature of the feature point based on the third-order polynomial fitting formula of curve fitting and performs secondary verification, sends the calibration result passing the secondary verification to the cloud server and the calibrated third-order polynomial fitting formula of curve fitting to the device end, and records the result not passing the secondary verification locally and reports to the cloud at the same time.

[0041] Further, the secondary verification of the edge gateway needs to meet the following two rules at the same time, specifically:

[0042] Rule 1: Review the temperature calibration process edge gateway receiving equipment end reported historical data whether in the detection index of the established range, if the data value is not in the established range, it is considered as failed verification;

[0043] Rule 2: Calculate the residual and standard deviation using the third-order polynomial fitting formula of curve fitting, if the residual and standard deviation exceed the interval range set by the cloud server, it is considered as failed verification.

[0044] The beneficial effects of the present application are:

[0045] The NTC temperature sensor calibration method based on edge computing of the present application can greatly shorten the period required for calibration compared with the traditional calibration method.

[0046] The NTC temperature sensor calibration method based on edge computing of the present application, the main process of calibration is carried out at the edge end, the equipment end does not need to send all the parameters to the cloud end, effectively saving the bandwidth consumption of the cloud server, reducing the computing pressure of the cloud end.

[0047] The NTC temperature sensor calibration method based on edge computing of the present application can effectively improve the precision and accuracy of the NTC temperature sensor. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0049] The specific embodiments of the present application will be further described in detail below in combination with the drawings.

[0050] As shown in Figure 1 The present application proposes an NTC temperature sensor calibration method based on edge computing, first, the edge node obtains the real-time data of the equipment end NTC temperature sensor through TCP / IP protocol, selects the optimal feature point for calibration based on genetic algorithm, and performs temperature data fitting analysis based on Steinhart-Hart algorithm at the feature point, secondly, the calibrated temperature model is updated to the equipment end through TCP / IP protocol, and the calibration result is sent to the cloud server, realizing the north-south communication of the edge gateway, completing the calibration of the temperature sensor at the edge end. The specific steps are as follows:

[0051] Step 1, deploy an edge computing network at the calibration site where the equipment calibration is needed.

[0052] Specifically, the edge computing network is composed of edge computing nodes and device nodes. The deployment of edge nodes utilizes container technology to build edge computing nodes in different types of devices, and the carriers of the edge nodes can be Windows, MacOS or Linux devices, etc. Each temperature sensor is a device node with wireless communication capability, which forms an edge computing network with the edge node through a specific communication protocol.

[0053] The number of edge computing nodes can be one or more according to actual needs.

[0054] Specifically, assuming that the number of devices to be calibrated is large and the time urgency is large, therefore, if the single edge computing node is under computing pressure, the number of edge computing nodes can be dynamically increased, and the parallel operation of multiple edge computing nodes can be performed to improve the calibration efficiency.

[0055] Step 2, the device end includes one or more standard devices and a plurality of devices to be calibrated, the device to be calibrated is an NTC temperature sensor to be calibrated, and the device end can collect temperature data containing detection indexes reported to the edge node during calibration, each device end sends the collected temperature data containing detection indexes to the edge computing node at a predetermined time interval.

[0056] For example, the device end sends the collected temperature, resistance value, time stamp, device number and environmental temperature to the edge node through the TCP / IP protocol, wherein the device number and environmental temperature only need to be sent once before the calibration process starts, and the temperature, resistance value and time stamp are uploaded to the edge node at a frequency of 5 seconds.

[0057] Step 3, the edge node analyzes the reported json format data and temporarily stores it in the local redis database, clears the local database data after one week, releases the storage space of the edge node, and facilitates short-term tracing. Specifically, the devices in the database use device numbers as primary keys to distinguish temperature, resistance and other data reported by different devices. The device uses whether the calibration is completed as a status bit, and when the device completes the calibration, the database automatically deletes the data under the device number to ensure the cleanliness of the database; the data of the device end that has not completed the calibration is retained in the database for 24 hours, and the database automatically cleans up the data after the time limit is exceeded.

[0058] Step 4, the edge node selects the best feature point for calibration through a genetic algorithm.

[0059] Specifically, for mass-produced temperature measurement modules, due to the differences between individual sensors, the input-output characteristics have obvious nonlinearity, and the individual differences are relatively large. When calibrating, if each feature point is calibrated, the time cost will be greatly increased. Therefore, when selecting feature points for fitting, there is a certain constraint relationship between the effect and the selected data. Therefore, the genetic algorithm is used to select feature points without changing the fitting effect, so as to obtain a solution with high efficiency, low cost and good fitting effect. The specific process is as follows:

[0060] (4.1) Establishment of gene and selection of feature point: Abstract each temperature data sent by the device end as a gene, define x q qth test point is a feature point, when x q =1 indicates that the qth test point is selected as a feature point, otherwise when x q =0 indicates that the qth test point is not selected as a feature point.

[0061] (4.2) Definition of fitness function: Based on the given optimization target, the fitness function is used to measure the quality of chromosomes in the population; Since the goal of decision is to reduce the total time cost, the fitness function is selected to measure the total time cost C total of chromosomes,

[0062] (4.2.1) Single measurement time cost: The time cost required to measure a temperature value is denoted as C q =50;

[0063] (4.2.2) Calibration error time cost:

[0064]

[0065] Symbol c j The time cost value corresponding to the single-point error. Where i represents the sensor number, j represents the test point, T i,j represents the actual temperature value, represents the sensor reading;

[0066] The calibration error time cost of a single sample body is calculated by

[0067] (4.2.3) Sample individual calibration time cost: The calibration time cost C i of the sample individual is the sum of the single determination time cost and the error calibration time cost;

[0068] C i =C s +C q ·n i ​

[0069] n i represents the number of measurement points in the calibration process of the sample individual;

[0070] (4.2.4) Total cost: Total cost C total is the average of the calibration time costs of all individuals, M is the total number of samples;

[0071]

[0072] (4.3) Cross operation: Calculate the offspring c1 and c2 from the parents p1 and p2 using simulated binary crossover, Calculate the parameters of the offspring by selecting a random number u in [0, 1) The offspring can be represented as

[0073] (4.4) Mutation operation: For the simulated genetic process, if a mutation occurs in an individual, each feature of the individual is multiplied by a random number in the range [1-m range , 1+m range ], m range is calculated independently as a multiplication factor according to each feature, according to the degree of convergence, m range ∈ [0.01, 0.3, 10% of the best individuals are selected from the current population by roulette selection method to enter the next generation iteration, so that the best individuals can be retained to the maximum extent;

[0074] (4.5) Termination condition: Genetic iteration is terminated when the following range is exceeded:

[0075]

[0076] σ 2 (r i ) is the variance of the i-th temperature feature point r i , σ 2 (W i ) is the variance of the i-th feature point combination W i , cov(r i -r j ) is the feature point i, j covariance, is the minimum input temperature, is the maximum input temperature.

[0077] For example, when the selected temperature feature points are 6, the corresponding total cost C totalThe cost is 350±10, and the time taken is 350±100s; when there are 5 feature points, the corresponding cost is 260±10, and the time taken is 600±100s. The increase in time is because the genetic algorithm iterates more when selecting feature points. The total cost is basically close to the number of feature points selected multiplied by the time cost per measurement.

[0078] Step 5: Using the temperature feature points obtained from the genetic algorithm, fit the temperature-resistance curves of the device to be calibrated and the standard at these feature points. The measured temperature at the feature points of the temperature sensor is accurately calculated using the fitted temperature model. The third-order polynomial fitting formula of the Steinhart-Hart algorithm is selected for fitting and analyzing the temperature data, as detailed below:

[0079] The third-order polynomial fitting formula for curve fitting of the RT meter data of the NTC thermistor is as follows: Where T is the open temperature scale in K, R is the resistance in KΩ, and A, B, and C are constant coefficients; the sum of squared errors is analyzed. That is:

[0080]

[0081] In the formula, R i For the i-th resistor, T i For R i Matching temperature value; constants use a i, Substituting and substituting into the polynomial fitting formula, we obtain the constant matrix:

[0082]

[0083] The constant coefficients of the fitting formula can be determined as follows:

[0084] A=(a23^2*b1-a12*a23*b3+a13*a22*b3-a13*a23*b2+a12*a33*b2-a22*a3 3*b1) / (a33*a12^2-2*a12*a13*a23+a22*a13^2+a11*a23^2-a11*a22*a33)

[0085] B=(a13^2*b2-a12*a13*b3+a11*a23*b3-a13*a23*b1-a11*a33*b2+a12*a3 3*b1) / (a33*a12^2-2*a12*a13*a23+a22*a13^2+a11*a23^2-a11*a22*a33)

[0086] C = (a12^2*b3-a12*a13*b2-a11*a22*b3+a11*a23*b2-a12*a23*b1+a13*a22*b1) / (a33*a12^2-2*a12*a13*a23+a22*a13^2+a11*a23^2-a11*a22*a33)

[0087] Thus, the temperature standard value at a certain resistance value is calculated.

[0088] The edge gateway calculates the temperature of the feature point based on the third-order polynomial fitting formula of curve fitting and performs secondary verification, sends the calibration result passing the secondary verification to the cloud server, and sends the calibrated third-order polynomial fitting formula of curve fitting to the device end, and records the result not passing the secondary verification locally and reports to the cloud at the same time.

[0089] The secondary verification of the edge gateway needs to meet the following two rules at the same time, specifically:

[0090] Rule 1: Review the temperature calibration process to check whether the historical data reported by the device end is within the specified range of the detection index. If the data value is not within the specified range, it is considered as not passing the verification.

[0091] Rule 2: Calculate the residual error and standard deviation using the third-order polynomial fitting formula of curve fitting. If the residual error and standard deviation exceed the interval range set by the cloud server, it is considered as not passing the verification. The operating system used in this example is Linux, the development language used on the edge side is go1.17, and the development language used on the device side is C language. Raspberry Pi 4B is used as the edge node, and the memory is 4GB. The data reported by the device end is fitted by selecting the following 6 points through genetic algorithm. The results show that after calibration, the residual error is controlled within ±1mK, the standard deviation is 0.0049℃, and the calibration effect is good.

[0092] Table 1 Data analysis before and after calibration

[0093]

[0094] Step 6: The edge node sends the calibration data and calibration result to the cloud to realize cloud storage of calibration records by using the MQTT protocol, and sends the feature function to the device end to complete the calibration by using the TCP / IP protocol.

[0095] The above examples are used to explain and illustrate the present application, but not to limit the present application. Any modifications and changes made to the present application within the spirit and protection scope of the claims fall within the protection scope of the present application.

Claims

1. An edge computing based NTC temperature sensor calibration method, characterized in that, The method comprises the following steps: Step 1: deploying an edge computing network at a calibration site; Step 2: after the calibration process starts, the device end sends the temperature data of the NTC temperature sensor to be calibrated and the standard device containing the detection index to the edge node through the TCP / IP protocol; the detection index contained is specifically resistance, temperature, timestamp, device number and environmental temperature; Step 3: the edge node analyzes the received data Step 4: the edge node selects the best calibration feature point through a genetic algorithm; in order to reduce the time cost of calibration, the genetic algorithm is used to select the optimal feature point for calibration, and the specific steps of obtaining the optimal feature point are as follows: (4.1) Gene establishment and selection of feature points: each temperature data sent by the device end is abstracted as a gene, and x q represents whether the qth test point is a feature point. When x q =1, it indicates that the qth test point is selected as a feature point, otherwise, when x q =0, it indicates that the qth test point is not selected as a feature point. (4.2) Definition of fitness function: based on the given optimization objective, the fitness function is used to measure the quality of chromosomes in the population; since the goal of the decision is to reduce the total time cost, the fitness function is selected as the reciprocal of the square of the total time cost C of the chromosome total ​ (4.2.1) Single measurement time cost: the time cost needed to measure one temperature value, denoted as C q ; (4.2.2) calibration error time cost: Symbol c j Time cost value corresponding to single point error; where i denotes the sensor number, j denotes the test point, T i,j denotes the actual temperature value, denotes the sensor reading; The calibration error time cost of a single sample body is used calculations; (4.2.3) Sample individual calibration time cost: the calibration time cost C of the sample individual i is the sum of the monomer calibration time cost and the error calibration time cost; C i = C s + C q • n i n i represents the number of measurement points in the individual calibration of the sample; (4.2.4) Total Cost: Total cost C total The average of the labeled time costs for all individuals, M is the total number of samples; (4.3) Cross operation: Calculate offspring cl and c2 from parents pl and p2 using simulated binary crossover; Calculate offspring parameters by selecting a random number u in [0, 1) and Offspring can be represented as (4.4) Mutation operation: To simulate the genetic process, if a individual is to be mutated, each feature of the individual is multiplied by a random number in the range [1-m range , 1+m range ], m range is calculated independently for each feature as a multiplication factor, m range ∈ [0.01, 0.3] according to the degree of convergence, 10% of the best individuals in the current population are selected by roulette selection method to enter the next generation of iterations, so that the best individuals can be retained to the maximum extent; (4.5) termination condition: if the following range is exceeded, the genetic iteration is terminated: σ 2 (r i ) is the variance of the i-th temperature feature point r i σ 2 (W i ) is the variance of the i-th feature point combination W i cov(r i -r j ) is the feature point i, j covariance, is the minimum input temperature, is the maximum input temperature; Step 5: according to the feature points obtained in step 4, the NTC temperature sensor to be calibrated and the standard device are calibrated at the feature points, and the measured temperature of the temperature sensor feature point is accurately calculated through the calibration data; the three-order polynomial fitting formula of the Steinhart-Hart algorithm is selected for fitting analysis of the temperature data, and the specific steps are as follows: The third-order polynomial fitting formula for curve fitting of the RT table data of NTC thermistor is selected: Wherein T is the open temperature scale, unit K, R is the resistance, unit KΩ, A, B, C is constant coefficient; by analyzing the error square sum That is: where R i is the i-th resistance, T i is the R i matched temperature value; constants use a i,j Substitute and bring into the polynomial fitting formula, get the constant matrix: The constant coefficient of the fitting formula is known: A = (a 23 ^2 * b1 - a 12 * a 23 * b3 + a 13 * a 22 * b3 - a 13 * a 23 * b2 + a 12 * a 33 * b2 - a 22 * a 33 * b1) / (a 33 * a 12 ^2 - 2 * a 12 * a 13 * a 23 + a 22 * a 13 ^2 + a 11 * a 23 ^2 - a 11 * a 22 * a 33 ) B = (a 13 ^2 * b2 - a 12 * a 13 * b3 + a 11 * a 23 * b3 - a 13 * a 23 * b1 - a 11 * a 33 * b2 + a 12 * a 33 * b1 ) / (a 33 * a 12 ^2 - 2 * a 12 * a 13 * a 23 + a 22 * a 13 ^2 + a 11 * a 23 ^2 - a 11 * a 22 * a 33 ) C = (a 12 ^2*b3 - a 12 *a 13 *b2 - a 11 *a 22 *b3 + a 11 *a 23 *b2 - a 12 *a 23 *b1 + a 13 *a 22 *b1) / (a 33 *a 12 ^2 - 2*a 12 *a 13 *a 23 +a 22 *a 13 ^2 + a 11 *a 23 ^2 - a 11 *a 22 *a 33 ) Thus, the temperature standard value at a certain resistance value is calculated; The edge gateway calculates the temperature of the feature point based on the three-order polynomial fitting formula of the curve fitting and performs secondary verification, and the edge gateway secondary verification needs to meet the following two rules at the same time, which are specifically: Rule 1: review whether the historical data reported by the device end and received by the edge gateway in the temperature calibration process is within the specified range of the detection index; if the data value is not within the specified range, it is considered as failed verification; Rule 2: calculate the residual error and standard deviation using the three-order polynomial fitting formula of the curve fitting; if the residual error and standard deviation exceed the interval range set by the cloud server, it is considered as failed verification; The calibration results that pass the secondary verification are sent to the cloud server, and the three-order polynomial fitting formula of the curve fitting after calibration is sent to the device end; the results that do not pass the secondary verification are recorded locally and reported to the cloud at the same time; Step 6: the edge node sends the calibration data and calibration results northward to the cloud to realize cloud storage of the calibration record, and sends the characteristic function southward to the device end through the TCP / IP protocol to complete the calibration.

Images