CN-121997086-A - Sensor product classification method under multi-mode condition

CN121997086ACN 121997086 ACN121997086 ACN 121997086ACN-121997086-A

Abstract

The invention discloses a sensor product classification method based on multi-modal conditions, which relates to the technical field of product classification and comprises the steps of defining the multi-modal conditions and standard values of a sensor, performing preprocessing operation on multi-dimensional measurement characteristic data of the sensor, screening effective samples to obtain a standardized deviation rate matrix, constructing multi-dimensional similarity components aiming at any two effective samples, setting similarity component weights, weighting and fusing to obtain comprehensive similarity, converting the comprehensive similarity into distance metric, clustering the effective samples by adopting a DBSCAN algorithm, outputting a clustering result and realizing the classification operation of the sensor product. According to the invention, the classification result can be consistent with the actual performance of the sensor through the multidimensional similarity component, the similarity component weight is supported to be adjusted according to the service requirement, classification targets of different application scenes can be adapted, abnormal samples which deviate from the standard seriously are eliminated by screening effective samples, the difference between the dimension and the base line is eliminated by standardization, and the interference of noise on the clustering result is reduced.

Inventors

LIU HAO
MA XIAOHUI
CHU FEI
XIAO WENLAN
ZHANG NAN

Assignees

山东国创微纳制造研究院有限公司
明石创新(烟台)微纳传感技术研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. The method for classifying the sensor products under the multi-mode condition is characterized by comprising the following steps of: Step S1, defining a multi-mode condition of a sensor and standard values under each mode condition to obtain multi-dimensional measurement characteristic data corresponding to the sensor under the multi-mode condition; S2, preprocessing operation is carried out on multidimensional measurement characteristic data of each sensor, and effective samples are screened to obtain a standardized deviation rate matrix of the effective samples; Step S3, constructing multidimensional similarity components among samples according to any two effective samples; s4, setting the weight of each similarity component according to the service requirement, and obtaining the comprehensive similarity of the corresponding sample through weighted fusion; s5, converting the comprehensive similarity into distance metric, clustering the effective samples by using a DBSCAN algorithm, outputting a clustering result, and realizing the classification operation of the sensor products according to the clustering result; The multi-dimensional similarity component comprises a deviation numerical value similarity component, a deviation trend similarity component, a standard fit similarity component and a standard trend matching similarity component.
2. The method for classifying sensor products according to claim 1, wherein said step S1 comprises: Defining a multimodal condition of a sensor as a set of temperatures And pressure set Obtaining 9-dimensional measurement characteristic data corresponding to the sensor under the multi-mode condition: , wherein, For a 9-dimensional measurement of the characteristics, Indicating that the sensor is at temperature Pressure and force The characteristic value under the condition that the characteristic value is, , ; Defining the standard value matrix under the mode condition as , wherein, Is a matrix of standard values, and is a matrix of standard values, Indicated at the corresponding temperature Pressure and force Standard values under the conditions.
3. The method for classifying sensor products according to claim 2, wherein said step S2 comprises: Step S21, calculating relative deviation rate for each characteristic value Calculate it and standard value Relative deviation ratio of (2) The formula is: ; s22, screening effective samples, namely setting the maximum allowable deviation rate Calculating the average relative deviation rate of each sensor The formula is: Reserved, reserve The sensor of (2) is used as a valid sample, and the rest samples are marked as abnormal samples; s23, performing Z-score standardization processing on the relative deviation rate of the 9-dimensional measurement characteristic data of the effective sample to obtain a standardized deviation rate matrix To eliminate dimensional effects.
4. A method for classifying a sensor product under a multi-modal condition according to claim 3, wherein the construction process of the bias numerical similarity component specifically comprises: Based on the reciprocal calculation of the normalized Euclidean distance, the consistency of any two effective samples A, B in the relative deviation rate is measured, and the expression of the similarity component of the deviation value comprises: ; ; Wherein, the For a normalized euclidean distance of the valid samples A, B, And The normalized bias rate matrices for the valid samples A, B respectively, , , In order to deviate the numerical similarity component, The value range is , Normalized euclidean distance maxima for all valid sample pairs.
5. The method for classifying sensor products under multi-modal conditions according to claim 4, wherein the construction process of the bias trend similarity component specifically comprises: Based on the pearson correlation coefficient calculation, the consistency of the deviation of any two effective samples A, B on the change trend of the deviation along with the pressure at the same temperature is measured, and the temperature is measured Extracting a sequence of pressure dimension deviations of the valid samples A, B And (3) with Expressed as: , Calculation of And (3) with Pearson correlation coefficient of (c) By the formula Conversion to And in the interval, eliminating the influence of negative correlation, taking the mean value of the pearson correlation coefficients after conversion at 3 temperatures as a deviation trend similarity component, wherein the expression is as follows: , wherein, 、 And The pearson correlation coefficients at 3 temperatures respectively, In order to deviate from the trend similarity component, The value range is 。
6. The method for classifying sensor products under multi-modal conditions according to claim 5, wherein the construction process of the standard fitness similarity component specifically comprises: Based on the overall fit calculation, the consistency of any two effective samples A, B on the overall fit with a standard value is measured, and the expression of the similarity component of the standard fit comprises: ; ; Wherein, the For the overall fit of each sample, To correspond to the average relative deviation rate of the samples, To maximum allowable deviation rate, ensure The value range is , To be an overall fit of valid sample a, To be an overall fit of the valid sample B, As a standard-fit similarity component, The value range is 。
7. The method for classifying sensor products under multi-modal conditions according to claim 6, wherein the construction process of the standard trend matching similarity component specifically comprises: measuring the consistency of any two effective samples A, B in the degree of trend matching with the standard value; For each temperature Judging the standard value sequence Trends include increasing, decreasing and fluctuating, and symbolizing the trend, resulting in a trend symbol; For each temperature Determining a normalized bias rate sequence of valid samples A, B Obtaining a standard trend symbol; Calculating trend matching rate of single sample The expression is: , for the amount of temperature for which the trend symbol corresponds to the standard trend symbol, The value range is ; Calculating a standard trend matching similarity component according to the trend matching rate difference complement value of the effective sample, wherein the expression is as follows: , wherein, For the trend match rate of the valid sample a, For the trend match rate of the valid sample B, For a standard trend to match the similarity component, The value range is 。
8. The method for classifying sensor products according to claim 7, wherein the expression of the integrated similarity in step S4 is: ; Wherein, the In order to integrate the degree of similarity, For the weight of the deviation value, For the bias trend weight, As the weight of the standard fitness, Weights are matched for standard trends.
9. The method for classifying sensor products according to claim 1, wherein said step S5 comprises: converting the comprehensive similarity into distance metric, clustering the effective samples by adopting a DBSCAN algorithm, and defining clustering parameters, wherein the clustering parameters comprise clustering radius And minimum density point number MinPts; The cluster radius The value of (2) is , For the minimum comprehensive similarity acceptable by service requirements, ensure that the comprehensive similarity is not less than "Valid samples are classified as density reachable; the value of the minimum density point MinPts is 1/8~1/10 of the number of the effective samples, so that false clustering of the isolated samples is avoided; and outputting a clustering result, and realizing the classification operation of the sensor products according to the clustering result.
10. The method according to claim 1, wherein the clustering result in the step S5 includes an effective cluster number, each cluster center, each cluster size, a clustered sensor index, and each clustered performance characteristic.

Description

Sensor product classification method under multi-mode condition Technical Field The invention relates to the technical field of product classification, in particular to a sensor product classification method under a multi-mode condition. Background In industrial applications, sensor products (e.g., pressure sensors, temperature sensors) often need performance testing under multi-modal conditions (e.g., different temperatures T 1-T3, different pressures P 1-P3 combined conditions) to obtain multi-dimensional measurement data (e.g., measurements corresponding to 9-dimensional characteristics ：T1P1、T1P2、T1P3、T2P1、T2P2、T2P3、T3P1、T3P2、T3P3). The classification of the sensor products needs to consider the consistency of the numerical value and the consistency of the condition response mode, and the degree of fit between the products and the standard needs to be evaluated by combining a preset standard value (namely a theoretical standard value under the condition of each mode), so that the classification result can reflect the actual use performance of the products. The existing sensor classification method has some technical defects: 1. Only the numerical similarity is concerned, the conditional response mode is ignored, the traditional method (such as clustering based on Euclidean distance and Manhattan distance) only calculates the numerical difference of multi-dimensional measurement data, the response trend (such as the change trend of a measurement value increasing with pressure at the same temperature) of the sensor under different conditions cannot be captured, the sensor with the 'numerical approach but opposite trend' is misclassified, such as the sensor A rises with the measurement value increasing with pressure at the temperature T 1, the sensor B falls with the measurement value increasing with pressure at the temperature T 1, the numerical approaches of the sensor A and the sensor B are classified as the same type, and the numerical approaches of the sensor A and the sensor B are inconsistent with the actual performance; 2. The standard value is not effectively fused for carrying out the fit evaluation, when the standard value is introduced, the existing method only simply calculates the absolute deviation between the measured value and the standard value, and the integral fit (namely the integral fit degree of the sensor and the standard value) is not incorporated into the classification basis, so that the unqualified sensor with small local deviation and low integral fit degree is misjudged as a qualified class; 3. The similarity calculation method has poor flexibility, the existing method mostly adopts equal weight average or single distance metric to calculate the similarity, and the importance of each evaluation dimension cannot be adjusted according to the service requirement (such as 'priority to ensure the fitness' or 'priority to ensure the trend consistency'), so that the suitability is insufficient. Based on the above, a multi-mode sensor product classification method is provided, which can eliminate the defects existing in the prior art scheme. Disclosure of Invention The invention aims to provide a multi-mode sensor product classification method, which aims to solve the problems that the existing method in the background art cannot give consideration to the response trend, the numerical deviation and the standard fitness of the multi-mode condition, so that the classification precision is low and the suitability is poor. In order to achieve the above purpose, the present invention provides the following technical solutions: a method for classifying sensor products under multi-mode conditions specifically comprises the following steps: Step S1, defining a multi-mode condition of a sensor and standard values under each mode condition to obtain multi-dimensional measurement characteristic data corresponding to the sensor under the multi-mode condition; S2, preprocessing operation is carried out on multidimensional measurement characteristic data of each sensor, and effective samples are screened to obtain a standardized deviation rate matrix of the effective samples; Step S3, constructing multidimensional similarity components among samples according to any two effective samples; s4, setting the weight of each similarity component according to the service requirement, and obtaining the comprehensive similarity of the corresponding sample through weighted fusion; s5, converting the comprehensive similarity into distance metric, clustering the effective samples by using a DBSCAN algorithm, outputting a clustering result, and realizing the classification operation of the sensor products according to the clustering result; The multi-dimensional similarity component comprises a deviation numerical value similarity component, a deviation trend similarity component, a standard fit similarity component and a standard trend matching similarity component. Further, the step S1 specifically includes: