Machine learning based pressure sensor test data processing method and system
By employing a machine learning-based pressure sensor test data processing method, the problem of detecting the dynamic response characteristics of sensors under complex working conditions was solved, achieving efficient and reliable quality assessment, and making it suitable for intelligent analysis and calibration in industrial settings.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHERN SCI & TECH (BEIJING) TECH DEV CO LTD
- Filing Date
- 2026-02-12
- Publication Date
- 2026-06-02
AI Technical Summary
Existing pressure sensor detection technologies struggle to reflect dynamic response characteristics under complex operating conditions and lack the integration of environmental coupling effects and historical fault data, resulting in biased quality assessments, high misjudgment rates, and an inability to meet the demands of high-end manufacturing for highly reliable sensors.
A machine learning-based pressure sensor test data processing method is adopted. By correcting the pressure display value, a dynamic pressure test curve is established, feature vectors are extracted, a temperature-pressure coupling model is constructed, a response consistency index is constructed, quality classification is performed, and the test environment is updated based on historical results.
It improves the accuracy and reliability of pressure sensor quality testing, significantly enhances testing standards in complex environments, and ensures the reliability of industrial applications.
Smart Images
Figure CN122132880A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of quality data detection technology, and in particular to a method and system for processing pressure sensor test data based on machine learning. Background Technology
[0002] In the field of industrial automation and intelligent equipment, pressure sensors are key sensing components, and their performance directly affects the safety and control accuracy of the system. Traditional pressure sensor quality testing mostly relies on static calibration methods, which only test the consistency of its output with the standard pressure value under ideal conditions such as constant temperature and no vibration. This makes it difficult to fully reflect the dynamic response characteristics of the sensor under complex operating conditions.
[0003] However, in real-world applications, pressure sensors often face multiple interference factors such as temperature fluctuations, mechanical vibrations, and rapid pressure changes, which can easily lead to problems such as temperature drift, hysteresis, overshoot, or response delay. Existing detection technologies lack the ability to perform refined modeling and analysis of dynamic response processes and fail to effectively integrate environmental coupling effects with historical fault data, resulting in biased quality assessments, high misjudgment rates, and an inability to meet the stringent requirements of high-end manufacturing for highly reliable sensors. Summary of the Invention
[0004] The purpose of this invention is to provide a pressure sensor test data processing method and system based on machine learning, so as to solve at least one of the problems existing in the prior art.
[0005] To achieve the above objectives, according to one aspect of this application, the present invention provides a machine learning-based method for processing pressure sensor test data, comprising:
[0006] The pressure display value is calibrated based on the test data, configuration parameters, and sampling frequency of the pressure sensor.
[0007] A dynamic pressure test curve is established based on the corrected pressure display value. The feature vector of the pressure sensor is extracted based on the dynamic pressure test curve. A temperature-pressure coupling model is established based on the test data. The feature vector of the pressure sensor is supplemented with the feature vector of the pressure sensor using the temperature-pressure coupling model.
[0008] A response consistency index is constructed based on the feature vector of the pressure sensor, and then the response state of the pressure sensor is determined based on the response consistency index.
[0009] Pressure sensors are classified according to their response status.
[0010] Update the test environment based on historical quality classification results.
[0011] Optionally, a temperature gradient constant t is set, and the temperature control range of the temperature control chamber is divided according to the temperature gradient constant t to obtain each test temperature point;
[0012] Set up a set of stepped pressure change rates and a set of scene vibrations for each test temperature point;
[0013] Using the set of temperature gradient constants, step pressure change rates, and scene vibration sets as a simulation scenario, a high-precision digital pressure controller is used to test the pressure sensor to obtain test data.
[0014] Optionally, the sampling time difference st between the pressure controller and the pressure sensor is calculated, and st is set to |1 / v1-1 / v2|, where v1 is the sampling frequency of the test data and v2 is the sampling frequency of the pressure sensor.
[0015] The product of the sampling time difference st and the pressure change rate at the pressure application time stamp t1 is used as the correction amount, and then the pressure display value P is subtracted from the correction amount to obtain the corrected pressure display value.
[0016] Optionally, a planar point plot is established using the corrected pressure display value and the pressure sensor value display timestamp t2, and the planar point plot is smoothed to obtain a dynamic pressure test curve;
[0017] The number of oscillations, settling time, and peak overshoot are extracted from the dynamic stress test curve. The extracted results are used as the reference value s(i) of the feature vector, where i is the number of the feature vector subvectors. Then, each subvector is compared with a preset value S(i). When s(i) is less than or equal to S(i), the i-th feature vector is set to 0. Otherwise, the absolute value of the offset ratio of s(i) relative to S(i) is used as the i-th feature vector.
[0018] Optionally, a temperature and pressure compensation coefficient Pcor is constructed;
[0019] The temperature and pressure compensation coefficient is compared with the coupling threshold. When the temperature and pressure compensation coefficient is greater than the coupling threshold, the absolute value of the offset ratio of Pcor relative to the coupling threshold is used as a supplementary feature vector and merged into the feature vector of the pressure sensor. Otherwise, no feature supplementation is performed on the feature vector of the pressure sensor.
[0020] Optionally, historical failure records of pressure sensors are collected, and the historical failure records are classified by feature vector subvectors. The classification results are then statistically analyzed to obtain the detection weight of each subvector.
[0021] The product of the detection weight of each sub-vector and the sub-vector of the feature vector is used as the response vector, and the magnitude of the response vector is used as the response consistency index.
[0022] Optionally, a response threshold is set, and the response consistency index is compared with the response threshold. If the response threshold is less than the response consistency index, the response state of the pressure sensor is determined to be normal; otherwise, the response state of the pressure sensor is determined to be abnormal.
[0023] Optionally, the response consistency index and response state at different test temperature points, test step pressure change rates, test vibration frequencies and test amplitudes can be stored;
[0024] Using the test temperature point, the test step pressure change rate, the test vibration frequency, and the test amplitude as disturbance terms, clustering is performed on the response consistency index with abnormal response states to obtain abnormal data clusters.
[0025] The number of each abnormal data cluster is counted and the result is denoted as N(j). Then, N(j) is compared with the preset number of distortion tests BN. If N(j) is less than BN, the abnormal data cluster is determined to be a distorted data cluster; if N(j) is greater than or equal to BN, the abnormal data cluster is determined to be a faulty data cluster; j is the number of abnormal data clusters.
[0026] If a fault data cluster exists in the pressure sensor, the pressure sensor is deemed unqualified in quality classification; if no fault data cluster exists in the pressure sensor, the pressure sensor is deemed qualified in quality classification.
[0027] Optionally, the fault clusters of the pressure sensors are statistically analyzed, and the interference terms within the fault clusters are further subdivided into intervals to obtain the updated test environment:
[0028] The number of times the test variable corresponding to the interference term within the fault cluster is counted and the statistical result is recorded as N(i)(k). When N(i)(k) exceeds BN, the k-th test variable of the i-th interference term is updated to N(i)(k)×[1-(N(i)(k)-BN) / BN] and N(i)(k)×[1+(N(i)(k)-BN) / BN]; otherwise, no update is performed.
[0029] According to another aspect of this application, a pressure sensor test data processing system based on machine learning is provided, comprising:
[0030] The data calibration module is used to calibrate the pressure display value based on the test data, configuration parameters and sampling frequency of the pressure sensor.
[0031] The feature construction module is used to establish a dynamic pressure test curve based on the corrected pressure display value, extract the feature vector of the pressure sensor based on the dynamic pressure test curve, establish a temperature-pressure coupling model based on the test data, and supplement the feature vector of the pressure sensor with the temperature-pressure coupling model.
[0032] The response analysis module is used to construct a response consistency index based on the feature vector of the pressure sensor, and then determine the response state of the pressure sensor based on the response consistency index.
[0033] The quality classification module is used to classify the pressure sensors according to their response status.
[0034] The environment update module is used to update the test environment based on historical quality classification results.
[0035] Compared with existing technologies, the advantages of this invention are as follows: This invention, through a machine learning-based pressure sensor test data processing method and system, effectively improves the efficiency of pressure sensor quality inspection. The system can accurately correct pressure display values, establish dynamic pressure test curves and extract feature vectors, supplement features through a temperature-pressure coupling model, and construct a response consistency index to accurately determine the sensor's response state, thus achieving quality classification. Simultaneously, the system can dynamically update the test environment based on historical quality classification results, significantly improving the quality inspection standards and factory quality of pressure sensors in complex environments, providing reliable assurance for industrial applications, and solving the problem of insufficient dynamic response characteristic analysis in existing technologies. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a flowchart illustrating the pressure sensor test data processing method based on machine learning in this embodiment.
[0038] Figure 2 This is a flowchart illustrating the feature vector construction method in this embodiment.
[0039] Figure 3 This is a flowchart illustrating the response status determination method in this embodiment.
[0040] Figure 4 This is a schematic diagram of the structure of the machine learning-based pressure sensor test data processing system provided in this embodiment. Detailed Implementation
[0041] To more clearly illustrate the present invention, the following description, in conjunction with preferred embodiments and accompanying drawings, further clarifies the invention. Similar components in the drawings are indicated by the same reference numerals. Those skilled in the art should understand that the specific description below is illustrative rather than restrictive and should not be construed as limiting the scope of protection of the present invention.
[0042] It should be noted that although the terms first, second, third, etc., may be used in the embodiments of this application for description, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of this application, first can also be referred to as second, and similarly, second can also be referred to as first.
[0043] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.
[0044] Specifically, the machine learning-based pressure sensor test data processing system provided in this application is used for: factory testing of pressure sensors, specifically for intelligent analysis and calibration of the dynamic response characteristics of pressure sensors in complex industrial environments, and for testing and verifying the sensor output data to obtain the quality test results of pressure sensor products.
[0045] In the above application scenarios, the quality monitoring of pressure sensors is not limited to detecting the relationship between static pressure and output. It is also necessary to analyze the dynamic response of the sensor under different environmental conditions and pressure change rates in order to improve the quality inspection standards and thus improve the factory quality of pressure sensors.
[0046] Specifically, before implementing the machine learning-based pressure sensor test data processing method and system, it is necessary to set up the test environment to collect test data and collect the configuration parameters of the pressure sensor.
[0047] The process of setting up the test environment is as follows:
[0048] Set the temperature gradient constant t, and divide the temperature control range of the temperature control chamber according to the temperature gradient constant t to obtain each test temperature point;
[0049] Set up a set of stepped pressure change rates and a set of scene vibrations for each test temperature point;
[0050] Using the set of temperature gradient constants, step pressure change rates, and scene vibration sets as a simulation scenario, a high-precision digital pressure controller is used to test the pressure sensor to obtain test data.
[0051] Specifically, the set of stepped pressure change rates in this application is [slow change rate, medium change rate, and fast change rate]. The set of stepped pressure change rates in this application refers to the change rate of the test pressure. The values of the slow change rate, medium change rate, and fast change rate are 0.1 MPa / s, 0.3 MPa / s, and 0.5 MPa / s, respectively. The set of scene vibrations in this application is a set of preset test vibration frequencies and preset test amplitudes. The set of scene vibrations includes multiple preset test vibration frequencies and preset test amplitudes. The values of the preset test vibration frequencies and preset test amplitudes are set by the user according to the common scenarios of the pressure sensor, and are not limited in this application.
[0052] Specifically, the temperature gradient constant t mentioned in this application is 5℃; the test data mentioned in this application includes: the pressure display value P of the pressure sensor, power supply noise interference data, temperature drift data, pressure application timestamp t1, pressure sensor value display timestamp t2, test temperature point, test step pressure change rate, test vibration frequency and test amplitude.
[0053] It is worth noting that, in this application, when testing the pressure sensor to obtain the test data of the pressure sensor, the test temperature points are repeatedly measured 5 times under the set of stepped pressure change rates and the set of scene vibrations, and the average value is taken after removing the maximum and minimum values as the test data of each test temperature point under the set of stepped pressure change rates.
[0054] The configuration parameters include: sampling frequency.
[0055] To apply the above-mentioned application scenarios, this application provides a pressure sensor test data processing method based on machine learning, the flowchart of which can be found in the document. Figure 1 As shown, it includes:
[0056] Step S101: Correct the pressure display value based on the test data, configuration parameters and sampling frequency of the pressure sensor.
[0057] Specifically, in step S101, the process of calibrating the pressure display value is as follows:
[0058] Calculate the sampling time difference st between the pressure controller and the pressure sensor, and set st=|1 / v1-1 / v2|, where v1 is the sampling frequency of the test data and v2 is the sampling frequency of the pressure sensor;
[0059] The product of the sampling time difference st and the pressure change rate at the pressure application time stamp t1 is used as the correction amount, and then the pressure display value P is subtracted from the correction amount to obtain the corrected pressure display value.
[0060] Specifically, by calculating the sampling time difference between the pressure controller and the sensor, and combining this with the pressure change rate to dynamically correct the displayed value, measurement deviations caused by asynchronous sampling are effectively eliminated. Compared to traditional static calibration methods, this approach significantly improves the real-time accuracy of pressure readings, providing a high-fidelity data foundation for subsequent feature extraction and status assessment.
[0061] Please continue reading. Figure 1 As shown, the machine learning-based pressure sensor test data processing method further includes:
[0062] Step S102: Establish a dynamic pressure test curve based on the corrected pressure display value, extract the feature vector of the pressure sensor based on the dynamic pressure test curve, establish a temperature-pressure coupling model based on the test data, and supplement the feature vector of the pressure sensor with the temperature-pressure coupling model.
[0063] Specifically, by establishing dynamic pressure test curves and extracting key dynamic response indicators such as oscillation count, settling time, and peak overshoot, a structured feature vector is formed. Simultaneously, a temperature-pressure coupling model is introduced to quantitatively compensate for the impact of temperature drift. This step enables a refined characterization of the sensor's multidimensional performance, enhancing the feature's ability to represent actual operating conditions.
[0064] Please see Figure 2 The diagram shown is a flowchart illustrating the feature vector construction method provided in this application, including:
[0065] Step S201: Establish a dynamic pressure test curve based on the calibrated pressure display value, and extract the feature vector of the pressure sensor based on the dynamic pressure test curve.
[0066] Specifically, in step S201, the process of extracting the feature vector is as follows:
[0067] A planar point plot is established using the corrected pressure display value and the pressure sensor value display timestamp t2, and the planar point plot is smoothed to obtain a dynamic pressure test curve.
[0068] The number of oscillations, settling time, and peak overshoot are extracted from the dynamic stress test curve. The extracted results are used as the reference value s(i) of the feature vector, where i is the number of the feature vector subvectors. Then, each subvector is compared with a preset value S(i). When s(i) is less than or equal to S(i), the i-th feature vector is set to 0. Otherwise, the absolute value of the offset ratio of s(i) relative to S(i) is used as the i-th feature vector.
[0069] Specifically, in this application, the baseline value of the component vector for the settling time is set to (1 / settling time) to ensure the consistency of the monotonicity of the data; in this application, the number of component vectors of the feature vector is not specifically limited, including but not limited to the number of oscillations, settling time, and peak overshoot; at the same time, the value of each preset value S(i) in this application is set by the user according to the test standard, and there is no specific limitation on its value in this application.
[0070] Please continue reading. Figure 2 As shown, the feature vector construction method further includes:
[0071] Step S202: Establish a temperature-pressure coupling model based on the test data, and use the temperature-pressure coupling model to supplement the feature vector of the pressure sensor.
[0072] Specifically, in step S202, the process of feature supplementation of the feature vector is as follows:
[0073] Construct a temperature and pressure compensation coefficient Pcor, and set Pcor=|P-Pd| / Pd×T0 / |T-T0|; where Pd is the ideal pressure value output by the pressure sensor at the same pressure and reference temperature, T is the current test temperature, and T0 is the reference temperature;
[0074] The temperature and pressure compensation coefficient is compared with the coupling threshold. When the temperature and pressure compensation coefficient is greater than the coupling threshold, the absolute value of the offset ratio of Pcor relative to the coupling threshold is used as a supplementary feature vector and merged into the feature vector of the pressure sensor. Otherwise, no feature supplementation is performed on the feature vector of the pressure sensor.
[0075] Specifically, the coupling threshold value mentioned in this application is 0.02; the reference temperature T0 is 25℃.
[0076] Please continue reading. Figure 1 As shown, the machine learning-based pressure sensor test data processing method further includes:
[0077] Step S103: Construct a response consistency index based on the feature vector of the pressure sensor, and then determine the response state of the pressure sensor based on the response consistency index.
[0078] Specifically, based on historical fault records, each feature vector is assigned a detection weight, a weighted response vector is constructed, and its magnitude is calculated as a response consistency index. This ensures that state determination not only relies on current test data but also incorporates prior experience. This method improves the sensitivity and reliability of anomaly identification and avoids the one-sidedness of judging by a single threshold.
[0079] Please see Figure 3 As shown, it is a flowchart illustrating the method for determining the response status of this application, including:
[0080] Step S301: Construct a response consistency index based on the feature vector of the pressure sensor.
[0081] Historical fault records of pressure sensors are collected, and these records are classified using feature vector subvectors. The classification results are then statistically analyzed to obtain the detection weights of each subvector.
[0082] The product of the detection weight of each sub-vector and the sub-vector of the feature vector is used as the response vector, and the magnitude of the response vector is used as the response consistency index.
[0083] Please continue reading. Figure 3 As shown, the response state determination method further includes:
[0084] Step S302: Determine the response state of the pressure sensor based on the response consistency index.
[0085] Set a response threshold and compare the response consistency index with the response threshold. If the response threshold is less than the response consistency index, the response state of the pressure sensor is determined to be normal; otherwise, the response state of the pressure sensor is determined to be abnormal.
[0086] Specifically, the response threshold value described in this application is 0.8.
[0087] Please continue reading. Figure 1 As shown, the machine learning-based pressure sensor test data processing method further includes:
[0088] Step S104: Classify the pressure sensors according to their response status.
[0089] Specifically, in step S104, the process of mass distribution for the pressure sensor is as follows:
[0090] The response consistency index and response state are stored for different test temperature points, test step pressure change rates, test vibration frequencies and test amplitudes;
[0091] Using the test temperature point, the test step pressure change rate, the test vibration frequency, and the test amplitude as disturbance terms, clustering is performed on the response consistency index with abnormal response states to obtain abnormal data clusters.
[0092] The number of each abnormal data cluster is counted and the result is denoted as N(j). Then, N(j) is compared with the preset number of distortion tests BN. If N(j) is less than BN, the abnormal data cluster is determined to be a distorted data cluster; if N(j) is greater than or equal to BN, the abnormal data cluster is determined to be a faulty data cluster; j is the number of abnormal data clusters.
[0093] If a fault data cluster exists in the pressure sensor, the pressure sensor is deemed unqualified in quality classification; if no fault data cluster exists in the pressure sensor, the pressure sensor is deemed qualified in quality classification.
[0094] Specifically, the value of the preset distortion test quantity BN mentioned in this application is set to 1% of the number of dynamic pressure test curves of the pressure sensor.
[0095] Specifically, by clustering abnormal response data and differentiating between distortions and genuine faults based on the number of clusters, misjudgments caused by occasional interference are effectively filtered out. This strategy enables intelligent grading of pressure sensor quality, improving the engineering practicality and robustness of the classification results while ensuring rigorous detection.
[0096] Please continue reading. Figure 1 As shown, the machine learning-based pressure sensor test data processing method further includes:
[0097] Step S105: Update the test environment based on the historical quality classification results. The process is as follows:
[0098] The fault clusters of pressure sensors are statistically analyzed, and the interference terms within the fault clusters are further subdivided into intervals to obtain the updated test environment:
[0099] The number of occurrences of the test variables corresponding to the interference terms within the fault cluster is counted and the result is denoted as N(i)(k). When N(i)(k) exceeds BN, the k-th test variable of the i-th interference term is updated to N(i)(k)×[1-(N(i)(k)-BN) / BN] and N(i)(k)×[1+(N(i)(k)-BN) / BN]; otherwise, no update is performed. Here, N(i)(k) represents the k-th test variable of the i-th interference term.
[0100] Specifically, the process of updating the test environment in this application is as follows: a certain test variable is divided into two test variables. For example, when the number of times the fault cluster of the pressure sensor occurs (30 times) at 50℃ exceeds BN(27), the test environment of 50℃ is updated to two test environments: 50×[1-(30-27) / 27]℃ and 50×[1+(30-27) / 27]℃.
[0101] Specifically, the test environment range is dynamically refined based on frequently occurring test variables in historical fault clusters, making the test conditions closer to actual failure scenarios. This mechanism achieves a self-optimizing closed loop in the testing system, continuously improving the targeting and foresight of the detection system, and helping to expose potential quality problems in advance.
[0102] Please see Figure 4The diagram shown is a structural schematic of the machine learning-based pressure sensor test data processing system provided in this application, including:
[0103] The data calibration module is used to calibrate the pressure display value based on the test data, configuration parameters and sampling frequency of the pressure sensor.
[0104] The feature construction module is used to establish a dynamic pressure test curve based on the corrected pressure display value, extract the feature vector of the pressure sensor based on the dynamic pressure test curve, establish a temperature-pressure coupling model based on the test data, and supplement the feature vector of the pressure sensor with the temperature-pressure coupling model.
[0105] The response analysis module is used to construct a response consistency index based on the feature vector of the pressure sensor, and then determine the response state of the pressure sensor based on the response consistency index.
[0106] The quality classification module is used to classify the pressure sensors according to their response status.
[0107] The environment update module is used to update the test environment based on historical quality classification results.
[0108] The machine learning-based pressure sensor test data processing system provided in this application can execute the machine learning-based pressure sensor test data processing method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of the method execution.
[0109] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.
Claims
1. A method for processing pressure sensor test data based on machine learning, characterized in that, include: The pressure display value is calibrated based on the test data, configuration parameters, and sampling frequency of the pressure sensor. A dynamic pressure test curve is established based on the corrected pressure display value. The feature vector of the pressure sensor is extracted based on the dynamic pressure test curve. A temperature-pressure coupling model is established based on the test data. The feature vector of the pressure sensor is supplemented with the feature vector of the pressure sensor using the temperature-pressure coupling model. A response consistency index is constructed based on the feature vector of the pressure sensor, and then the response state of the pressure sensor is determined based on the response consistency index. Pressure sensors are classified according to their response status. Update the test environment based on historical quality classification results.
2. The pressure sensor test data processing method based on machine learning according to claim 1, characterized in that, Set the temperature gradient constant t, and divide the temperature control range of the temperature control chamber according to the temperature gradient constant t to obtain each test temperature point; Set up a set of stepped pressure change rates and a set of scene vibrations for each test temperature point; Using the set of temperature gradient constants, step pressure change rates, and scene vibration sets as a simulation scenario, a high-precision digital pressure controller is used to test the pressure sensor to obtain test data.
3. The pressure sensor test data processing method based on machine learning according to claim 2, characterized in that, Calculate the sampling time difference st between the pressure controller and the pressure sensor, and set st=|1 / v1-1 / v2|, where v1 is the sampling frequency of the test data and v2 is the sampling frequency of the pressure sensor; The product of the sampling time difference st and the pressure change rate at the pressure application time stamp t1 is used as the correction amount, and then the pressure display value P is subtracted from the correction amount to obtain the corrected pressure display value.
4. The pressure sensor test data processing method based on machine learning according to claim 3, characterized in that, A planar point plot is established using the corrected pressure display value and the pressure sensor value display timestamp t2, and the planar point plot is smoothed to obtain a dynamic pressure test curve. The number of oscillations, settling time, and peak overshoot are extracted from the dynamic stress test curve. The extracted results are used as the reference value s(i) of the feature vector, where i is the number of the feature vector subvectors. Then, each subvector is compared with a preset value S(i). When s(i) is less than or equal to S(i), the i-th feature vector is set to 0. Otherwise, the absolute value of the offset ratio of s(i) relative to S(i) is used as the i-th feature vector.
5. The pressure sensor test data processing method based on machine learning according to claim 4, characterized in that, Construct the temperature and pressure compensation coefficient Pcor; The temperature and pressure compensation coefficient is compared with the coupling threshold. When the temperature and pressure compensation coefficient is greater than the coupling threshold, the absolute value of the offset ratio of Pcor relative to the coupling threshold is used as a supplementary feature vector and merged into the feature vector of the pressure sensor. Otherwise, no feature supplementation is performed on the feature vector of the pressure sensor.
6. The pressure sensor test data processing method based on machine learning according to claim 5, characterized in that, Historical fault records of pressure sensors are collected, and these records are classified using feature vector subvectors. The classification results are then statistically analyzed to obtain the detection weights of each subvector. The product of the detection weight of each sub-vector and the sub-vector of the feature vector is used as the response vector, and the magnitude of the response vector is used as the response consistency index.
7. The pressure sensor test data processing method based on machine learning according to claim 6, characterized in that, Set a response threshold and compare the response consistency index with the response threshold. If the response threshold is less than the response consistency index, the response state of the pressure sensor is determined to be normal; otherwise, the response state of the pressure sensor is determined to be abnormal.
8. The pressure sensor test data processing method based on machine learning according to claim 7, characterized in that, The response consistency index and response state are stored for different test temperature points, test step pressure change rates, test vibration frequencies and test amplitudes; Using the test temperature point, the test step pressure change rate, the test vibration frequency, and the test amplitude as disturbance terms, clustering is performed on the response consistency index with abnormal response states to obtain abnormal data clusters. The number of each abnormal data cluster is counted and the result is recorded as N(j). Then, N(j) is compared with the preset number of distortion tests BN. When N(j) is less than BN, the abnormal data cluster is determined to be a distorted data cluster; when N(j) is greater than or equal to BN, the abnormal data cluster is determined to be a faulty data cluster. j represents the number of clusters of abnormal data; If a fault data cluster exists in the pressure sensor, the pressure sensor is deemed unqualified in quality classification; if no fault data cluster exists in the pressure sensor, the pressure sensor is deemed qualified in quality classification.
9. The pressure sensor test data processing method based on machine learning according to claim 8, characterized in that, The fault clusters of pressure sensors are statistically analyzed, and the interference terms within the fault clusters are further subdivided into intervals to obtain the updated test environment: The number of times the test variable corresponding to the interference term within the fault cluster is counted and the statistical result is recorded as N(i)(k). When N(i)(k) exceeds BN, the k-th test variable of the i-th interference term is updated to N(i)(k)×[1-(N(i)(k)-BN) / BN] and N(i)(k)×[1+(N(i)(k)-BN) / BN]; otherwise, no update is performed.
10. A pressure sensor test data processing system based on machine learning, applied to the pressure sensor test data processing method based on machine learning as described in any one of claims 1-9, characterized in that, include: The data correction module is used to correct the pressure display value based on the test data, configuration parameters and sampling frequency of the pressure sensor. The feature construction module is used to establish a dynamic pressure test curve based on the corrected pressure display value, extract the feature vector of the pressure sensor based on the dynamic pressure test curve, establish a temperature-pressure coupling model based on the test data, and supplement the feature vector of the pressure sensor with the temperature-pressure coupling model. The response analysis module is used to construct a response consistency index based on the feature vector of the pressure sensor, and then determine the response state of the pressure sensor based on the response consistency index. The quality classification module is used to classify the pressure sensors according to their response status. The environment update module is used to update the test environment based on historical quality classification results.