Multi-modal condition sensor product classification method

Abstract

The invention discloses a sensor product classification method based on a multi-modal condition, and relates to the technical field of product classification, and the method comprises the steps: defining the multi-modal condition and a standard value of a sensor; performing preprocessing operation on the multi-dimensional measurement characteristic data of the sensor, and screening effective samples to obtain a standardized deviation rate matrix; constructing a multi-dimensional similarity component for any two effective samples; setting a similarity component weight, and performing weighted fusion to obtain a comprehensive similarity; and the comprehensive similarity is converted into distance metric, effective samples are clustered by adopting a DBSCAN algorithm, a clustering result is output, and classification operation of sensor products is realized. Through the multi-dimensional similarity component, the classification result can be consistent with the actual performance of the sensor, the weight of the similarity component can be adjusted according to business requirements, classification targets of different application scenes can be adapted, abnormal samples seriously deviating from the standard are eliminated by screening effective samples, the difference between the dimension and the baseline is eliminated through standardization, and the classification accuracy is improved. And the interference of noise on the clustering result is reduced.

Description

Technical Field

[0001] This invention relates to the field of product classification technology, specifically a method for classifying sensor products under multimodal conditions. Background Technology

[0002] In industrial applications, sensor products (such as pressure sensors and temperature sensors) often require performance testing under multimodal conditions (such as combinations of different temperatures T1-T3 and different pressures P1-P3) to obtain multidimensional measurement data (such as the measured values ​​corresponding to 9-dimensional features: T1P1, T1P2, T1P3, T2P1, T2P2, T2P3, T3P1, T3P2, and T3P3). The classification of these sensor products must consider both numerical consistency and conditional response mode consistency, and must also incorporate preset standard values ​​(i.e., theoretical baseline values ​​under each modal condition) to assess the product's fit with the standard, ensuring that the classification results reflect the product's actual performance.

[0003] Existing sensor classification methods have some technical shortcomings: 1. Focusing only on numerical similarity and ignoring conditional response patterns, traditional methods (such as clustering based on Euclidean distance and Manhattan distance) only calculate the numerical differences of multidimensional measurement data and cannot capture the response trend of sensors under different conditions (such as the trend of measurement value change with increasing pressure at the same temperature). This leads to the misclassification of sensors with "similar values ​​but opposite trends". For example, sensor A's measurement value increases with increasing pressure at T1, while sensor B's measurement value decreases with increasing pressure at T1. The two have similar values ​​but are classified into the same category, which does not match the actual performance. 2. The existing method does not effectively integrate standard values ​​for fit assessment. When introducing standard values, it simply calculates the "absolute deviation between the measured value and the standard value" without including the "overall fit" (i.e. the overall degree of conformity between the sensor and the standard value) in the classification criteria. This results in unqualified sensors with "small local deviations but low overall fit" being misclassified as qualified. 3. The similarity calculation method is not flexible. Existing methods mostly use equal weighted average or a single distance metric to calculate similarity, which cannot adjust the importance of each evaluation dimension according to business needs (such as "prioritizing fit" or "prioritizing trend consistency"), resulting in insufficient adaptability.

[0004] Based on this, a multimodal sensor product classification method is now provided, which can eliminate the drawbacks of existing technical solutions. Summary of the Invention

[0005] The purpose of this invention is to provide a sensor product classification method under multimodal conditions, so as to solve the problem that existing methods in the background art cannot take into account the response trend, numerical deviation and standard fit under multimodal conditions, resulting in low classification accuracy and poor adaptability.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for classifying sensor products under multimodal conditions, specifically including the following steps: Step S1: Define the multimodal conditions of the sensor and the standard values ​​under each modal condition to obtain the multidimensional measurement characteristic data of the sensor under the multimodal conditions; Step S2: Perform preprocessing operations on the multidimensional measurement feature data of each sensor, screen valid samples, and obtain the standardized deviation rate matrix of the valid samples; Step S3: For any two valid samples, construct a multi-dimensional similarity component between the samples; Step S4: Set the weights of each similarity component according to business needs, and perform weighted fusion to obtain the comprehensive similarity of the corresponding samples; Step S5: Convert the comprehensive similarity into a distance metric, use the DBSCAN algorithm to cluster the valid samples, output the clustering results, and perform the classification operation of the sensor products based on the clustering results; The multi-dimensional similarity components include deviation numerical similarity components, deviation trend similarity components, standard fit similarity components, and standard trend matching similarity components.

[0007] Further, step S1 specifically includes: The multimodal conditions of the sensor are defined as a temperature set. With pressure set The 9-dimensional measurement feature data of the sensor under this multimodal condition were obtained: ,in, It is a 9-dimensional measurement feature. Indicates the sensor's temperature ,pressure Eigenvalues ​​under certain conditions , ; Define the standard value matrix under the above modal conditions as follows: ,in, For the standard value matrix, Indicates at the corresponding temperature ,pressure Standard values ​​under given conditions.

[0008] Further, step S2 specifically includes: Step S21: Calculate the relative deviation rate: for each feature value Calculate its value compared with the standard value. relative deviation rate The formula is: ; Step S22: Screening valid samples: Setting the maximum permissible deviation rate Calculate the average relative deviation rate for each sensor. The formula is: ,reserve The sensors are used as valid samples, and the remaining samples are marked as abnormal samples; Step S23: Perform Z-score standardization on the relative deviation rate of the 9-dimensional measurement feature data of the valid samples to obtain the standardized deviation rate matrix. To eliminate the influence of dimensions.

[0009] Furthermore, the construction process of the deviation numerical similarity component specifically includes: Based on the reciprocal calculation of standardized Euclidean distance, the similarity of relative deviation rates between any two valid samples A and B is measured. The expression for the numerical similarity component of the deviation includes: ; ; in, The standardized Euclidean distance between valid samples A and B. and These are the standardized deviation rate matrices for valid samples A and B, respectively. , , The deviation is a numerical similarity component. The range of values ​​is , This represents the maximum standardized Euclidean distance for all valid sample pairs.

[0010] Furthermore, the construction process of the deviation trend similarity component specifically includes: Based on the Pearson correlation coefficient, this measure assesses the consistency of the deviation trends of any two valid samples A and B with pressure at the same temperature. For each temperature... Extract the stress dimension deviation sequence of valid samples A and B. and , is represented as: , ,calculate and Pearson correlation coefficient Through formula Convert to The interval is used to eliminate the influence of negative correlation. Then, the mean of the transformed Pearson correlation coefficients at three temperatures is taken as the bias trend similarity component, expressed as: ,in, , and The Pearson correlation coefficients at the three temperatures are shown below. For the deviation trend similarity component, The range of values ​​is .

[0011] Furthermore, the construction process of the standard fit similarity component specifically includes: Based on the overall fit calculation, the consistency between any two valid samples A and B and the overall fit with the standard value is measured. The expression of the standard fit similarity component includes: ; ; in, The overall fit for each sample, This represents the average relative deviation rate of the corresponding samples. To ensure the maximum permissible deviation rate The range of values ​​is , For the overall fit of valid sample A, For the overall fit of valid sample B, For the standard fit similarity component, The range of values ​​is .

[0012] Furthermore, the construction process of the standard trend matching similarity component specifically includes: It measures the consistency between any two valid samples A and B in terms of their trend matching with the standard value; For each temperature Determine the standard value sequence The trend, which includes increasing, decreasing and fluctuating trends, is represented by symbols to obtain trend symbols; For each temperature Determine the standardized deviation rate sequence of valid samples A and B. The trend is used to obtain a standard trend symbol; Calculate the trend matching rate for a single sample The expression is: , The number of temperatures for which the trend symbol matches the standard trend symbol. The range of values ​​is ; The standard trend matching similarity component is calculated based on the difference in trend matching rate among valid samples. The expression is as follows: ,in, The trend matching rate of valid sample A. The trend matching rate of valid sample B. For standard trend matching similarity components, The range of values ​​is .

[0013] Furthermore, the expression for the comprehensive similarity in step S4 is: ; in, To assess overall similarity, Weights are the numerical values ​​of the deviation. For deviation trend weight, As the standard fit weight, Standard trend matching weights.

[0014] Furthermore, step S5 specifically includes: The comprehensive similarity is converted into a distance metric, and the DBSCAN algorithm is used to cluster the valid samples. Clustering parameters are defined, including the cluster radius. and the minimum density point number MinPts; The cluster radius The value is , To ensure that the minimum overall similarity is acceptable to business requirements, the overall similarity is ≥ The effective samples are classified as having a density that can be reached; The minimum density point number MinPts is set to 1 / 8 to 1 / 10 of the number of effective samples to avoid isolated samples forming false clusters. Output the clustering results and perform classification operations on sensor products based on the clustering results.

[0015] Furthermore, the clustering results in step S5 include the number of effective clusters, the cluster centers, the size of each cluster, the sensor index of the cluster, and the performance characteristics of each cluster.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention uses multi-dimensional similarity components to simultaneously consider numerical deviation, conditional response trend, standard fit, and trend-standard matching degree, avoiding misclassification such as "numerical similarities but opposite trends" and "small local deviations but overall unqualified", so that the classification results can be consistent with the actual performance of the sensor and improve accuracy. 2. This invention supports adjusting the weights of each similarity component according to business needs, which can adapt to the classification targets of different application scenarios. Moreover, the clustering results of this invention not only include the traditional "cluster size and index", but also output the performance characteristics of each cluster such as "average deviation rate", "overall fit" and "trend matching rate", which makes it easier to understand the physical meaning of clustering. 3. This invention filters out abnormal samples that deviate significantly from the standard by screening effective samples, eliminates differences between the units and the baseline by standardization, and reduces the interference of noise on the clustering results, thus showing good application prospects. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the steps of the present invention.

[0018] Figure 2 This is a schematic diagram of the process of the present invention.

[0019] Figure 3 This is a scatter plot of the clustering results of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0021] In this embodiment, as Figure 1 and Figure 2 As shown, a multimodal sensor product classification method is proposed. For the 9-dimensional measurement feature data of sensors under multimodal conditions (e.g., 3 temperatures × 3 pressures), the method first converts the measurement data into a "relative deviation rate" through preprocessing, then constructs multi-dimensional similarity components, calculates the comprehensive similarity using a weighted fusion strategy, and finally achieves accurate classification of sensor products based on density clustering (DBSCAN algorithm). This solves the technical problem that existing sensor product classification methods cannot simultaneously consider multimodal response trends, numerical deviations, and standard fit, resulting in low classification accuracy and poor adaptability. The method specifically includes the following steps: Step S1: Define the multimodal conditions of the sensor and the standard values ​​under each modal condition to obtain the multidimensional measurement characteristic data of the sensor under the multimodal conditions; Step S2: Perform preprocessing operations on the multidimensional measurement feature data of each sensor, screen valid samples, and obtain the standardized deviation rate matrix of the valid samples; Step S3: For any two valid samples, construct a multi-dimensional similarity component between the samples. The multi-dimensional similarity component includes a deviation numerical similarity component, a deviation trend similarity component, a standard fit similarity component, and a standard trend matching similarity component. Step S4: Set the weights of each similarity component according to business requirements, and perform weighted fusion to obtain the comprehensive similarity of the corresponding samples. The expression for the comprehensive similarity is: Step S5: Convert the comprehensive similarity into a distance metric, use the DBSCAN algorithm to cluster the valid samples, output the clustering results, and perform the classification operation of the sensor products based on the clustering results.

[0022] Specifically, such as Figure 2 As shown, step S1 includes: The multimodal conditions of the sensor are defined as a temperature set. With pressure set The 9-dimensional measurement feature data of the sensor under this multimodal condition were obtained: ,in, It is a 9-dimensional measurement feature. Indicates the sensor's temperature ,pressure Eigenvalues ​​under certain conditions , ; Define the standard value matrix under the above modal conditions as follows: ,in, For the standard value matrix, Indicates at the corresponding temperature ,pressure Standard values ​​under given conditions.

[0023] Specifically, such as Figure 2 As shown, this invention employs targeted data preprocessing optimization operations. It eliminates the difference in the magnitude of standard values ​​under multimodal conditions through "relative deviation rate" and removes severely deviated samples in advance by "average relative deviation rate screening". Then, it eliminates dimensional interference through standardization, thereby improving the robustness of subsequent clustering and solving the problem that existing methods are greatly affected by baseline differences and outliers. Step S2 includes: Step S21: Calculate the relative deviation rate: for each feature value Calculate its value compared with the standard value. relative deviation rate The formula is: ; Step S22: Screening valid samples: Setting the maximum permissible deviation rate (If set at 10% according to industry standards, this value can be adjusted according to actual needs.) Calculate the average relative deviation rate for each sensor. The formula is: ; reserve The sensors are used as valid samples, and the remaining samples are marked as abnormal samples and classified separately. This operation can exclude unqualified products that deviate significantly from the standard. Step S23: Perform Z-score standardization on the relative deviation rate of the 9-dimensional measurement feature data of the valid samples, with a mean of 0 and a standard deviation of 1, to obtain the standardized deviation rate matrix. Eliminate the influence of dimensions.

[0024] Specifically, such as Figure 2As shown, this invention constructs a multi-dimensional similarity component system, breaking through the limitation of traditional methods that only focus on "numerical similarity". It constructs four core components: "deviation value, deviation trend, standard fit, and standard trend matching", and incorporates "response pattern consistency under multimodal conditions" and "overall fit" into the similarity evaluation, thus solving the misclassification problem of "numerical similarity but opposite trends, and local compliance but overall deviation". Step S3 includes: The construction process of the deviation numerical similarity component specifically includes: calculating the reciprocal of the standardized Euclidean distance to measure the consistency of the relative deviation rates between any two valid samples A and B. The expression for the deviation numerical similarity component includes: , ,in, The standardized Euclidean distance between valid samples A and B. and These are the standardized deviation rate matrices for valid samples A and B, respectively. , , The deviation is a numerical similarity component. The range of values ​​is , The closer the value is to 1, the higher the consistency of the deviation magnitude. The maximum standardized Euclidean distance for all valid sample pairs; The construction process of the deviation trend similarity component specifically includes: calculating based on the Pearson correlation coefficient to measure the consistency of the deviation trends of any two valid samples A and B with pressure at the same temperature, and for each temperature... Extract the stress dimension deviation sequence of valid samples A and B. and , is represented as: , ,calculate and Pearson correlation coefficient Through formula Convert to The interval is used to eliminate the influence of negative correlation. Then, the mean of the transformed Pearson correlation coefficients at three temperatures is taken as the bias trend similarity component, expressed as: ,in, , and The Pearson correlation coefficients at the three temperatures are shown below. For the deviation trend similarity component, The range of values ​​is , The closer to 1, the higher the trend consistency; The construction process of the standard fit similarity component specifically includes: based on the overall fit calculation, measuring the consistency between any two valid samples A and B in terms of overall fit with the standard value. The expression of the standard fit similarity component includes: , ,in, The overall fit for each sample, This represents the average relative deviation rate of the corresponding samples. To ensure the maximum permissible deviation rate The range of values ​​is , The closer it is to 1, the higher the overall fit. For the overall fit of valid sample A, For the overall fit of valid sample B, For the standard fit similarity component, The range of values ​​is , The closer it is to 1, the higher the overall consistency and compatibility. The construction process of the standard trend matching similarity component specifically includes: measuring the consistency between any two valid samples A and B in terms of trend matching with the standard value, and for each temperature... Determine the standard value sequence The trend, including increasing, decreasing, and fluctuating trends, is represented by symbols, such as increasing = 1, decreasing = -1, and fluctuating = 0, resulting in a trend symbol for each temperature. Determine the standardized deviation rate sequence of valid samples A and B. The trend is observed, a standard trend sign is obtained, and the trend matching rate of a single sample is calculated. The expression is: , The number of temperatures for which the trend symbol matches the standard trend symbol. The range of values ​​is The standard trend matching similarity component is calculated based on the difference in trend matching rate among valid samples. The expression is as follows: ,in, The trend matching rate of valid sample A. The trend matching rate of valid sample B. For standard trend matching similarity components, The range of values ​​is , The closer to 1, the higher the consistency of the trend match.

[0025] Specifically, such as Figure 2As shown, this invention adopts a flexible weighted fusion strategy, which dynamically adjusts the weight of each similarity component based on business needs. For example, quality detection focuses on "standard fit", while mechanism analysis focuses on "deviation trend". Compared with existing equal weight or single distance methods, it can adapt to classification targets in different scenarios and avoid the defects of single evaluation. The expression for the comprehensive similarity in step S4 is: ; in, To assess overall similarity, , , and These are all weight parameters, and their setting range is: This represents the numerical weight of the deviation, ranging from 0.3 to 0.4, used to prioritize ensuring consistency in the magnitude of the deviation. Based on the weights, This is the deviation trend weight, ranging from 0.2 to 0.3, used to ensure trend consistency. As auxiliary weights, This is the standard fit weight, ranging from 0.25 to 0.35, used to ensure overall fit. As the core weight, The standard trend matching weight, ranging from 0.05 to 0.15, is used to assist in constraining the trend and matching the standard. As a secondary weight; In this embodiment, the weights are set as follows: , , , Ensure overall similarity The range of values ​​is , The closer the value is to 1, the higher the performance consistency between the two sensors.

[0026] Specifically, such as Figure 2 As shown, this invention uses interpretable clustering results output, which not only outputs the traditional number, size, and index of clusters, but also simultaneously provides the "average deviation rate, overall fit, trend matching rate" and performance characteristics of each cluster, linking the clustering results with the actual performance of the sensor, and solving the problems of abstract and difficult-to-implement results of existing methods; Step S5 includes: A metric, also known as a distance function, is a special function in a metric space that satisfies specific conditions. It is generally used in metric spaces. express; Comprehensive similarity Convert to distance metric (distance) , The smaller the value, the higher the similarity. The DBSCAN (Density-Based Spatial Clustering Applications and Noise) algorithm is used to cluster the valid samples. Clustering parameters are defined, including the cluster radius. Minimum density points (MinPts), cluster radius This ensures that the business-acceptable lower similarity limit is transformed into a distance upper limit in the clustering algorithm. The MinPts ratio is set empirically to prevent noise from forming small clusters and can be adjusted according to the data distribution. It is set as follows: Cluster radius The value is , The minimum overall similarity acceptable to business requirements, for example: ,but Ensure that the overall similarity is ≥ The effective samples are classified as having a density that can be reached; The minimum density point count, MinPts, is set to 1 / 8 to 1 / 10 of the number of valid samples. For example, if there are 100 valid samples, then... To avoid isolated samples forming false clusters; The output clustering results include: the number of effective clusters, the cluster centers (including deviation centers and fit centers), the size of each cluster, the sensor index of each cluster, and the performance characteristics of each cluster (such as "high fit and stable type" and "low fit and variability type"). Finally, the sensor products are classified based on the clustering results.

[0027] Specifically, the implementation process of the present invention will be further illustrated with the following examples: Step S1: Select 100 pressure sensors of the same model and test them under multimodal conditions ( , , , , , Nine-dimensional measurement data were collected under these conditions to obtain the nine-dimensional measurement feature data of the sensor under this multimodal condition: ; Define the standard value matrix under the above modal conditions as follows: The unit is mV; Step S2: Set the maximum permissible deviation rate before measurement. Calculate the relative deviation rate of 100 sensors, assuming a deviation of 10%. For example, the measurement value of sensor A is: ,but By analogy, the relative deviation rate matrix of A is obtained, and the average relative deviation rate of each sensor is calculated. Filter out There were 85 valid samples and 15 abnormal samples. The relative deviation rates of the 9-dimensional measurement feature data of 85 valid samples were Z-score standardized to obtain the standardized deviation rate matrix. ; Step S3: For all sample pairs of the 85 valid samples (a total of 85 × 84 / 2 = 3570 pairs), construct and calculate the multi-dimensional similarity component (deviation numerical similarity component) between any sample pair. Deviation trend similarity component Standard fit similarity component Matching similarity components with standard trends ); Step S4: Set before calculation , , , Then according to the weighted formula Calculate the overall similarity ; Step S5: Set the minimum overall similarity before conversion. A value of 0.75 corresponds to a cluster radius. The minimum density points (MinPts) are set to 0.25, and the minimum number of density points (MinPts) is 12 (1 / 8 of the effective sample size). The overall similarity is then calculated. Convert to distance This yields an 85×85 distance matrix. Clustering is then performed using the DBSCAN algorithm, with the distance matrix and cluster radius as input. The minimum density point count (MinPts) is used to perform DBSCAN clustering, resulting in 3 valid clusters (C1, C2, C3) and 8 noisy samples (isolated samples with no density reachable from the valid samples). Figure 3 As shown, Figure 3 This is a scatter plot of the clustering results for sensor samples. Based on PCA dimensionality reduction, the clustering output is reduced to 2 dimensions. C1, C2, and C3 are valid clusters. Abnormal samples and noise samples are marked with different colors to intuitively show the classification effect. Finally, the clustering results are output, including: Number of clusters: 3 valid clusters, 1 group of outlier samples, and 1 group of noisy samples; Cluster centers: C1: 30 samples, bias center = 0.03~0.1, average fit ≈92%, trend matching rate ≈100%; C2: 25 samples, bias center = 0.13~0.2, average fit ≈80%, trend matching rate ≈96%; C3: 22 samples, bias center = 0.26~0.35, average fit ≈72%, trend matching rate ≈85%; Clustering characteristics: C1 is a high-fit stable type (small deviation, trend is completely consistent with the standard, which is a core qualified product), C2 is a medium-fit stable type (medium deviation, trend is basically consistent, which is a regular qualified product), and C3 is a low-fit critical type (deviation is close to the threshold, some trends are abnormal, which need to be monitored for use). Abnormal samples: severely deviated ( (The fault needs to be investigated). Noise sample: Isolated type (the fit is acceptable but there are no similar samples, and it needs to be verified separately).

[0028] In summary, through actual working condition testing of the sensors in the clustering results, the sensors in the C1 cluster have better performance stability than those in the C2 and C3 clusters across the entire temperature-pressure range. All abnormal samples have hardware faults (such as diaphragm damage), which verifies the accuracy and practicality of the classification results of this invention.

[0029] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for classifying sensor products under multimodal conditions, characterized in that, Specifically, the following steps are included: Step S1: Define the multimodal conditions of the sensor and the standard values ​​under each modal condition to obtain the multidimensional measurement characteristic data of the sensor under the multimodal conditions; Step S2: Perform preprocessing operations on the multidimensional measurement feature data of each sensor, screen valid samples, and obtain the standardized deviation rate matrix of the valid samples; Step S3: For any two valid samples, construct a multi-dimensional similarity component between the samples; Step S4: Set the weights of each similarity component according to business needs, and perform weighted fusion to obtain the comprehensive similarity of the corresponding samples; Step S5: Convert the comprehensive similarity into a distance metric, use the DBSCAN algorithm to cluster the valid samples, output the clustering results, and perform the classification operation of the sensor products based on the clustering results; The multi-dimensional similarity components include deviation numerical similarity components, deviation trend similarity components, standard fit similarity components, and standard trend matching similarity components.

2. The sensor product classification method under multimodal conditions according to claim 1, characterized in that, Step S1 specifically includes: The multimodal conditions of the sensor are defined as a temperature set. With pressure set The 9-dimensional measurement feature data of the sensor under this multimodal condition were obtained: ,in, It is a 9-dimensional measurement feature. Indicates the sensor's temperature ,pressure Eigenvalues ​​under certain conditions , ; Define the standard value matrix under the above modal conditions as follows: ,in, For the standard value matrix, Indicates at the corresponding temperature ,pressure Standard values ​​under given conditions.

3. The sensor product classification method under multimodal conditions according to claim 2, characterized in that, Step S2 specifically includes: Step S21: Calculate the relative deviation rate: for each feature value Calculate its value compared with the standard value. relative deviation rate The formula is: ; Step S22: Screening valid samples: Setting the maximum permissible deviation rate Calculate the average relative deviation rate for each sensor. The formula is: ,reserve The sensors are used as valid samples, and the remaining samples are marked as abnormal samples; Step S23: Perform Z-score standardization on the relative deviation rate of the 9-dimensional measurement feature data of the valid samples to obtain the standardized deviation rate matrix. To eliminate the influence of dimensions.

4. The sensor product classification method under multimodal conditions according to claim 3, characterized in that, The construction process of the deviation numerical similarity components specifically includes: Based on the reciprocal calculation of standardized Euclidean distance, the similarity of relative deviation rates between any two valid samples A and B is measured. The expression for the numerical similarity component of the deviation includes: ； ； in, The standardized Euclidean distance between valid samples A and B. and These are the standardized deviation rate matrices for valid samples A and B, respectively. , , The deviation is a numerical similarity component. The range of values ​​is , This represents the maximum standardized Euclidean distance for all valid sample pairs.

5. The sensor product classification method under multimodal conditions according to claim 4, characterized in that, The construction process of the deviation trend similarity component specifically includes: Based on the Pearson correlation coefficient, this measure assesses the consistency of the deviation trends of any two valid samples A and B with pressure at the same temperature. For each temperature... Extract the stress dimension deviation sequence of valid samples A and B. and , is represented as: , ,calculate and Pearson correlation coefficient Through formula Convert to The interval is used to eliminate the influence of negative correlation. Then, the mean of the transformed Pearson correlation coefficients at three temperatures is taken as the bias trend similarity component, expressed as: ,in, , and The Pearson correlation coefficients at the three temperatures are shown below. For the deviation trend similarity component, The range of values ​​is .

6. The sensor product classification method under multimodal conditions according to claim 5, characterized in that, The construction process of the standard fit similarity components specifically includes: Based on the overall fit calculation, the consistency between any two valid samples A and B and the overall fit with the standard value is measured. The expression of the standard fit similarity component includes: ； ； in, The overall fit for each sample, This represents the average relative deviation rate of the corresponding samples. To ensure the maximum permissible deviation rate The range of values ​​is , For the overall fit of valid sample A, For the overall fit of valid sample B, For the standard fit similarity component, The range of values ​​is .

7. The sensor product classification method under multimodal conditions according to claim 6, characterized in that, The construction process of the standard trend matching similarity component specifically includes: It measures the consistency between any two valid samples A and B in terms of their trend matching with the standard value; For each temperature Determine the standard value sequence The trend, which includes increasing, decreasing and fluctuating trends, is represented by symbols to obtain trend symbols; For each temperature Determine the standardized deviation rate sequence of valid samples A and B. The trend is used to obtain a standard trend symbol; Calculate the trend matching rate for a single sample The expression is: , The number of temperatures for which the trend symbol matches the standard trend symbol. The range of values ​​is ; The standard trend matching similarity component is calculated based on the difference in trend matching rate among valid samples. The expression is as follows: ,in, The trend matching rate of valid sample A. The trend matching rate of valid sample B. For standard trend matching similarity components, The range of values ​​is .

8. The sensor product classification method under multimodal conditions according to claim 7, characterized in that, The expression for the comprehensive similarity in step S4 is: ； in, To assess overall similarity, Weights are the numerical values ​​of the deviation. For deviation trend weight, As the standard fit weight, Standard trend matching weights.

9. The sensor product classification method under multimodal conditions according to claim 1, characterized in that, Step S5 specifically includes: The comprehensive similarity is converted into a distance metric, and the DBSCAN algorithm is used to cluster the valid samples. Clustering parameters are defined, including the cluster radius. and the minimum density point number MinPts; The cluster radius The value is , To ensure that the minimum overall similarity is acceptable to business requirements, the overall similarity is ≥ The effective samples are classified as having a density that can be reached; The minimum density point number MinPts is set to 1 / 8 to 1 / 10 of the number of effective samples to avoid isolated samples forming false clusters. Output the clustering results and perform classification operations on sensor products based on the clustering results.

10. The sensor product classification method under multimodal conditions according to claim 1, characterized in that, The clustering results in step S5 include the number of effective clusters, the cluster centers, the size of each cluster, the sensor index of each cluster, and the performance characteristics of each cluster.

Images