CN117786549A - Speed reducer health state evaluation method, system, equipment and medium based on combined weighting method - Google Patents

Abstract

Speed reducer health state assessment method, system, equipment and medium based on combined weighting method, wherein the method comprises the following steps: signal acquisition, signal data processing, wireless transmission, classification processing, optimal characteristic component extraction, index standardization, subjective weight and objective weight determination, weight determination and health state assessment; the system, the device and the medium are used for realizing a speed reducer health state assessment method based on a combined weighting method; the invention utilizes the weighting method of the combination of the analytic hierarchy process and the entropy weighting method to reasonably allocate the weights, and then combines the gray clustering method and the fuzzy comprehensive evaluation method to analyze and determine the influence degree among the factors, thereby realizing the health state evaluation of the speed reducer, improving the accuracy and the service life of the health monitoring of the speed reducer, and greatly reducing the calculation cost of the computer detection.

Description

Speed reducer health state evaluation method, system, equipment and medium based on combined weighting method

Technical Field

The invention relates to the technical field of health monitoring of speed reducers, in particular to a speed reducer health state assessment method, a speed reducer health state assessment system, speed reducer health state assessment equipment and speed reducer health state assessment media based on a combined weighting method.

Background

With the rapid development of industry, mechanical devices, especially speed reducers, are increasingly paid attention to, and are widely applied to the fields of wind power generation, aviation and the like for regulating rotating speed and torque. However, due to the specificity of the industrial environment, the gears and bearings of the speed reducer often run for a long time in a high-speed and heavy-load environment, so that faults are easy to occur, huge economic losses can be brought to enterprises, and normal use of mechanical equipment can be influenced. Therefore, it is very important to effectively detect and pre-warn the real-time state of the speed reducer, and improve the reliability of the speed reducer. Common health assessment methods for gearboxes include wear analysis, vibration analysis, temperature analysis, and the like. The vibration analysis-based method is to utilize vibration signals to realize fault diagnosis of the gearbox. Faults of a speed reducer, a gear box, a bearing and the like can be reflected through vibration phenomena. The vibration signals which can reflect the equipment faults can be collected and extracted by selecting proper signal collecting sensors and different signal extracting modes aiming at the speed reducers in different industries. The low-frequency, medium-frequency and high-frequency vibration signals contain useful information of abnormal operation of equipment and various faults. By analyzing and processing the information, the health state of the speed reducer can be evaluated.

Patent application publication No. CN114357663B discloses a training gear box fault diagnosis model method and a gear box fault diagnosis method, wherein the acquired current signals are used for calculating characteristic values representing the complexity and mutation degree of the current signals; screening the characteristic values to generate a sample set; and training the reinforcement learning network model to generate a gear box fault diagnosis model. However, this patent application refers only to the current signal characteristics to perform gearbox fault diagnosis, and the characteristic dimension is low, so that the diagnosis accuracy cannot be ensured.

The patent application with the publication number of CN106586841A discloses a method and a system for monitoring the running state of a speed reducer of hoisting equipment, wherein the monitoring method comprises the following steps: collecting and storing acceleration signals of a bearing when the monitored speed reducer operates; and performing fault diagnosis on the monitored speed reducer according to the preset early warning value and the acceleration signal. And setting an early warning value for the acceleration signal, and executing warning when judging that the acceleration signal exceeds the early warning value. However, the patent application only uses a simple early warning value to perform fault diagnosis, which may cause erroneous judgment and inaccurate fault diagnosis.

In summary, in the prior art, the analysis of the original signal is mostly single analysis of the time domain or the frequency domain, the characteristic value is not comprehensive enough, and the data analysis is not comprehensive; using fourier transform analysis, the effect of the interference component cannot be considered; failure to consider actual conditions of the speed reducer fault during fault diagnosis and state detection; when the training model is used for fault diagnosis and state detection, the model has poor generalization capability and obvious parameter influence due to insufficient fault signal types, and is difficult to apply in a large scale in a factory.

The existing fault early warning threshold and state detection are easy to cause misjudgment due to the fact that the selected characteristic value is single and is influenced by noise and other interference components, and the detection result is not accurate enough.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention aims to provide a method for evaluating the health state of a speed reducer based on a combined weighting method, which reasonably distributes weights by using a weighting method combining a hierarchical analysis method and an entropy weighting method; then, combining a gray clustering method and a fuzzy comprehensive evaluation method, and analyzing and determining the influence degree among the factors; the invention realizes the evaluation of the health state of the speed reducer, improves the accuracy rate and the service life of the health monitoring of the speed reducer, and greatly reduces the calculation cost of the computer detection.

In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:

the method for evaluating the health state of the speed reducer based on the combined weighting method comprises the following steps:

step 1, signal acquisition: randomly collecting acceleration signals, rotation speed signals and current signals of speed reducers with different health states and different types;

step 2, signal data processing: on embedded edge hardware, carrying out noise reduction processing and characteristic value extraction on the acceleration signal and the rotating speed signal acquired in the step 1 to obtain a characteristic data set, wherein the characteristic data in the characteristic data set comprises a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic;

Step 3, wireless transmission: transmitting the characteristic data set obtained in the step 2 to a remote server database;

step 4, classification processing: the remote server classifies the speed reducers with different types and different working conditions according to the speed reducer types, the rotating speed signals and the current signals in the step 1 to obtain N types, wherein each type comprises a plurality of speed reducers of the type;

step 5, extracting optimal characteristic components: collecting normal sample data and fault sample data of each type of speed reducer in the step 4 respectively, selecting optimal characteristic components according to the time domain characteristic values, the frequency domain characteristic values and the characteristic value domain characteristic values extracted in the step 2, sequencing the extracted characteristics by using a filtering algorithm, determining an optimal subset by using a Binary Search (BS) and a Radial Basis (RBF) network, selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS), and finally comparing the two subsets to obtain the optimal subset with the highest recognition accuracy among the two subsets;

step 6, index standardization: performing index standardization processing on the data in the optimal subset of each type of speed reducer obtained in the step 5;

step 7, determining subjective weight and objective weight: respectively determining subjective weight and objective weight of the characteristic data indexes after the standardization in the step 6;

Step 8, determining the weight value: carrying out combined weighting on the subjective weight and the objective weight in the step 7, and determining a weight;

step 9, health status assessment: and (3) dividing the health state of the speed reducer by using a gray clustering method and combining the weight determined in the step (8).

In the step 1, an edge node is arranged on a bearing seat of an input shaft and a bearing seat of an output shaft of a speed reducer, and a high-precision acceleration sensor is integrated inside the edge node to collect acceleration signals; acquiring a rotating speed signal by externally connecting a photoelectric tachometer on an edge node of an input shaft bearing seat of a speed reducer; collecting current signals by externally connecting a current sensor on an independent edge node; the model of the high-precision acceleration sensor is 608A11, and the frequency range is 0.5000-10000Hz.

The extracting of the optimal characteristic component in the step 5 specifically comprises the following steps:

step 5.1: collecting characteristic data sets of each type of speed reducer under normal operation and under different fault conditions, and evaluating fault characteristics from distance measurement, information measurement and correlation measurement by using 4 filtering algorithms, namely a Fisher score method (FS), a Distance Evaluation Technology (DET), an information gain (MI) and a Pearson (Person) correlation coefficient, so as to obtain 4 groups of characteristic scores;

The calculation formula of the fischer score method (FS):

wherein x is ₊ 、x _- Respectively positive and negative samples, l ₊ 、l _- Respectively representing the number of positive samples and negative samples;

distance assessment technique (DET):

euclidean distance

Wherein q is _i Representing the average distance, p, of the optimal feature subset _i Representing an optimal feature subsetIs a distance between the features of (a) and (b);

the information gain (MI) is based on the information measure to evaluate the feature, is the measure of the difference between the prior uncertainty and the expected posterior uncertainty of the feature, and the calculation formula is as follows:

I(x；y)＝H(y)-H(y|x)；

wherein H (y) represents information entropy, and H (y|x) represents conditional entropy; giving a feature x and a label y corresponding to the feature x;

pearson (Person) correlation coefficient measures the correlation between features and classes, and the formula is calculated:

wherein x and y are respectively specific to a given feature and its corresponding label;

step 5.2: sorting the features according to the feature scores obtained in the step 5.1, selecting the top n feature input Radial Basis (RBF) networks with the highest rank to classify, then taking the recognition error rate as a weight value and taking the product of the recognition error rate and the rank of each feature as a new score of the corresponding feature, repeating the operation on 4 groups of feature scores to obtain 4 new scores for each feature, summing the new scores as a weighted score result of the feature, wherein a weighting mechanism is shown in the following formula, and finally reordering according to the sequence from small to large;

S _NEW ＝R _FS E _FS +R _DET E _DET +R _MI E _MI +R _PCC E _PCC ；

Wherein S is _NEW A weighted score representing a feature, R _FS And E is _FS Respectively represent the ranking and the recognition error rate of the characteristic of the Fisher Score (FS) model in the evaluation result of the corresponding model, R _DET And E is _DET Respectively represent the ranking and recognition error rate of the feature in the evaluation result of the corresponding model by the Distance Evaluation Technology (DET), R _MI And E is _MI Respectively representing the ranking and recognition error rate of the characteristic of the information gain (MI) in the evaluation result of the corresponding model, R _PCC And E is _PCC Respectively representRanking and identifying error rate of the feature in the evaluation result of the corresponding model;

step 5.3: and (3) rapidly screening redundant and irrelevant features after ranking by using a Binary Search (BS), determining an optimal subset, then selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS) respectively with the subset obtained by the Binary Search (BS) as a starting point, and finally comparing the two subsets to obtain the optimal subset with the one with the highest recognition accuracy.

The method for standardizing the index in the step 6 comprises the following steps:

the indexes in the index system are classified into larger and more optimal indexes and smaller and more optimal indexes;

the larger and more optimal index standardized expression is:

wherein: x is the index value after standardization; x is an index actual running value; xmax is an index running upper limit; xmin is an index operation lower limit value;

The smaller and more optimal index standardized expression is:

wherein: x is the index value after standardization; x is an index actual running value; x' max is an index operation upper limit value; x' min is the index operation lower limit value.

The subjective weight determination in the step 7 is specifically:

and 7.1.1, comparing the two indexes by two according to the mutual correlation influence by using an analytic hierarchy process to form a judgment matrix, wherein the judgment matrix A is defined as follows:

wherein A is a judgment matrix, and a is an element in the judgment matrix;

step 7.1.2, calculating a to-be-optimized transfer matrix e of the judgment matrix A so as to enable the to-be-optimized transfer matrix e to meet consistency test; converting the judgment matrix A into an antisymmetric matrix B, specifically:

B＝lgA；

wherein A is a judgment matrix, and B is an antisymmetric matrix;

step 7.1.3, obtaining an optimal transfer matrix C by means of solution, wherein the optimal transfer matrix C is specifically:

wherein C is an optimal transfer matrix, and b is an element in an antisymmetric matrix;

step 7.1.4, calculating a to-be-optimized transfer matrix e, which specifically comprises the following steps:

wherein e is a transmission matrix to be optimized, and C is an optimal transmission matrix;

step 7.1.5, according to the to-be-optimized transfer matrix e obtained in step 7.1.4, adopting a canonical column average method to obtain a feature vector W corresponding to the maximum feature value, namely a weight vector; the normalization processing for each column of elements is specifically as follows:

In the method, in the process of the invention,e is a transmission matrix to be optimized;

step 7.1.6, normalized judgment matrixAdding according to columns, specifically:

in the method, in the process of the invention,is a normalized judgment matrix, +.>Is a judgment matrix added according to columns;

step 7.1.7, for the judgment matrix added by columnNormalization processing is carried out to obtain subjective weight w of the characteristic data index _i The method specifically comprises the following steps:

wherein i=1, 2, 3..n, w _i Subjective weight w for characteristic data index _i ；

The objective weight determination in the step 7 is specifically:

step 7.2.1, performing standardized processing on the evaluation index data to obtain a dimensionless matrix R, wherein the dimensionless matrix R is specifically:

R＝(r _ij ) _m×n ；

wherein R is a dimensionless matrix, R is an element in the dimensionless matrix R, i is an ith index, and j is a jth index;

step 7.2.2, calculating the information entropy e of the ith index _i The method specifically comprises the following steps:

in the method, in the process of the invention,ei is the information entropy of the ith index, i is the ith index, j is the jth index, and R is an element in the dimensionless matrix R;

step 7.2.3, calculating the entropy vector of the ith index to obtain the objective weight v of the characteristic data index _i The method specifically comprises the following steps:

in the formula, v _i Objective weight of characteristic data index e _i The information entropy of the ith index, i being the ith index.

The weight determining in the step 8 is specifically:

subjective weight w using the characteristic data index obtained in step 7.1.7 _i And objective weight v of characteristic data index obtained in step 7.2.3 _i Combining with a combination weighting method to obtain comprehensive weights to obtain an index combination weight vector W _i The method comprises the following steps:

wherein w is _i Subjective weight of characteristic data index, v _i Objective weight of characteristic data index, W _i The weight vectors are combined for the indicators.

The health state evaluation in the step 9 specifically includes:

step 9.1, setting n evaluation systems and m evaluation indexes of an evaluation object, wherein the n evaluation systems and the m evaluation indexes are divided into s different gray classes; the index j (j=1, … m) of the system i (i=1, … n) has a sample value x _ij Dividing the system i into ash classes k (k=1, …, S), wherein the value range of the ash class k is as followsThen->And->Respectively a whitening weight function and a weight of the index j belonging to the gray class k;

step 9.2, calculating the membership degree corresponding to the actual value of the index, wherein the function expression is as follows:

in the method, in the process of the invention,a whitening weight function with index j belonging to gray class k, wherein k is gray class;

step 9.3, calculating the clustering coefficient of the index j on the ash class k according to the step 9.2The method comprises the following steps:

in the method, in the process of the invention,a whitening weight function with index j belonging to gray class k, k being gray class +.>Is a clustering coefficient;

Step 9.4, judging what ash class the system i belongs to under the index j, namely what health state the speed reducer is in, dividing according to a whitening weight function, and if the index j is in the range of (1.0-0.8), judging that the speed reducer is in a health state; if the index j is (0.8-0.6), the speed reducer is in a sub-health state; if the index j is within the range of (0.6-0.4), the speed reducer is in a fault state, and if the index j is within the range of (0.4-0), the speed reducer is in a scrapped state.

Speed reducer health state evaluation system based on combination weighting method includes:

the signal acquisition module: randomly collecting acceleration signals, rotation speed signals and current signals of speed reducers with different health states and different types;

a signal data processing module: on embedded edge hardware, carrying out noise reduction processing and characteristic value extraction on the acceleration signal and the rotating speed signal acquired by the signal acquisition module to obtain a characteristic data set, wherein the characteristic data in the characteristic data set comprises a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic;

and a wireless transmission module: transmitting the characteristic data set obtained by the signal data processing module to a remote server database;

the classification processing module: the remote server classifies the speed reducers with different types and different working conditions according to the speed reducer types, the rotating speed signals and the current signals in the signal acquisition module to obtain N types, wherein each type comprises a plurality of speed reducers of the type;

And an optimal characteristic component extraction module: the method comprises the steps of respectively collecting normal sample data and fault sample data of each type of speed reducer in a classification processing module, selecting optimal characteristic components according to a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic value extracted by a signal data processing module, sequencing the extracted characteristics by using a filtering algorithm, determining an optimal subset by using a Binary Search (BS) and a Radial Basis (RBF) network, selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS), and finally comparing the two subsets to obtain the optimal subset of each type, wherein the optimal subset has the highest recognition accuracy;

index standardization module: performing index standardization processing on the data in the optimal subset of each type of speed reducer obtained by the optimal characteristic component extraction module;

and the subjective weight and objective weight determining module is used for: respectively determining subjective weight and objective weight of the characteristic data index standardized by the index standardization module;

weight determining module: the subjective weight and the objective weight of the subjective weight and objective weight determining module are combined and weighted to determine a weight value;

health status assessment module: and dividing the health state of the speed reducer by using a gray clustering method and combining the weight determined by the weight determining module.

Speed reducer health state evaluation equipment based on combination weighting method includes:

a memory: a computer program for storing the speed reducer health state assessment method based on the combined weighting method;

a processor: the method is used for realizing the speed reducer health state evaluation method based on the combined weighting method when the computer program is executed.

A computer readable storage medium storing a computer program which, when executed by a processor, enables implementation of a method of assessing a health state of a speed reducer based on a combined weighting method.

Compared with the prior art, the invention has the beneficial effects that:

1. the analysis of the original acceleration signal in the step 2 is not limited to the independent analysis of the time domain or the frequency domain, the multiple analysis of the time domain, the frequency domain and the characteristic value domain is used, the frequency domain analysis relates to a low-frequency part and a high-frequency part, the characteristic value selection of the signal is more comprehensive, and compared with the prior art, the method is more accurate in fault identification and more accurate in determined weight.

2. The feature extraction algorithm in the step 2 is low in complexity, can be completed in most embedded hardware nodes in the mainstream of the current market, and has the characteristics of simplicity and low cost compared with the prior art.

3. In the invention, the acceleration signal is acquired by adopting high-precision sensor hardware in the step 1, the sampling rate reaches 10KHz, the acquired data is accurate, and the typical fault frequency range of the speed reducer can be covered; typical fault frequency of the speed reducer is distributed between 1 and 10KHZ, and the problems of large communication data volume and high price can be brought by utilizing the existing wired system, so that the speed reducer cannot be popularized on a large scale. The high-precision acceleration sensor is integrated in the edge node hardware, and the central processing unit in the edge node hardware is used for carrying out feature extraction calculation, so that the core features of acceleration signals are extracted.

4. The noise reduction algorithm in the step 2 greatly reduces the influence of noise and other interference components on the characteristic value, shortens the characteristic extraction time and greatly improves the quality of the selected characteristic.

5. The feature selection method provided by the step 2 greatly reduces the number of features, shortens the time of classification training and greatly improves the accuracy.

6. The invention adopts the whitening weight function and the combination weight mode to ensure that the obtained health state grade of the speed reducer is more accurate, thereby being beneficial to prolonging the service life of the speed reducer.

In summary, the invention aims to treat a system with incomplete information such as a speed reducer as a gray system, and reflects subjective and objective information by using a weighting method combined by a hierarchical analysis method and an entropy weighting method, so that the weight is more scientific, rigorous and reasonable; and then, by combining a gray clustering method and a fuzzy comprehensive evaluation method, the influence degree among all factors is analyzed and determined, and the health state evaluation of the speed reducer is realized.

Drawings

FIG. 1 is a flow chart of the method of the present invention.

Fig. 2 is a flow chart of the weight determination of the present invention.

Fig. 3 is a flow chart of the evaluation of the health status of the speed reducer according to the present invention.

Detailed Description

The invention is described in further detail below with reference to the accompanying drawings.

Referring to fig. 1, the method for evaluating the health state of the speed reducer based on the combined weighting method comprises the following steps:

step 1, signal acquisition: randomly collecting acceleration signals, rotation speed signals and current signals of speed reducers with different health states and different types; the method comprises the following steps: the high-precision acceleration sensor is integrated inside the edge node to collect acceleration signals by installing the edge node on the bearing seat of the input shaft and the bearing seat of the output shaft of the speed reducer; acquiring a rotating speed signal by externally connecting a photoelectric tachometer on an edge node of an input shaft bearing seat of a speed reducer; collecting current signals by externally connecting a current sensor on an independent edge node; the model of the high-precision acceleration sensor is 608A11, and the frequency range is 0.5000-10000Hz.

Step 2, signal data processing: on embedded edge hardware, carrying out noise reduction processing and characteristic value extraction on the acceleration signal and the rotating speed signal acquired in the step 1 to obtain a characteristic data set, wherein the characteristic data in the characteristic data set comprises a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic; the method comprises the steps of processing the acquired acceleration signals and the acquired rotation speed signals while transmitting the signals by the sensor, realizing noise reduction of vibration signals and extraction of characteristic values, greatly reducing communication data volume on the premise of keeping core characteristics required by a diagnosis algorithm, and simultaneously obtaining a characteristic data set; the method comprises the following steps:

on embedded edge hardware, the feature data set obtained by carrying out noise reduction and feature extraction calculation on the original acceleration signal and the rotating speed signal comprises a time domain feature value, a frequency domain feature value and a feature value domain feature, and specific parameters are shown in the following table 1:

table 1 time and frequency domain characteristic parameters

Characteristic value field characteristic parameter

In the table, x (n) represents the time of vibration signalA sequence of samples, N representing the number of samples of the time sequence, s (K) representing the spectrum of the signal x (N), k=1, 2 … K, K being the number of spectral lines, f (K) representing the frequency value of the kth spectral line, Characteristic value spectrum of the graph signal, characteristic value, i=1, 2, …, N.

Step 3, wireless transmission: transmitting the characteristic data set obtained in the step 2 to a remote server database;

step 4, classification processing: the remote server classifies the speed reducers with different types and different working conditions according to the speed reducer types, the rotating speed signals and the current signals in the step 1 to obtain N types, wherein each type comprises a plurality of speed reducers of the type;

step 5, extracting optimal characteristic components: collecting normal sample data and fault sample data of each type of speed reducer in the step 4 respectively, selecting optimal characteristic components according to the time domain characteristic values, the frequency domain characteristic values and the characteristic value domain characteristic values extracted in the step 2, sequencing the extracted characteristics by using a filtering algorithm, roughly determining an optimal subset by using Binary Search (BS) and Radial Basis (RBF) networks, selecting the optimal subset by using Sequence Forward Search (SFS) and Sequence Backward Search (SBS), and finally comparing the two subsets to obtain the optimal subset of each type, wherein the optimal subset with the highest recognition accuracy; the method comprises the following steps:

the extracting of the optimal characteristic component in the step 5 specifically comprises the following steps:

step 5.1: collecting characteristic data sets of each type of speed reducer under normal operation and under different fault conditions, and evaluating fault characteristics from distance measurement, information measurement and correlation measurement by using 4 filtering algorithms, namely a Fisher score method (FS), a Distance Evaluation Technology (DET), an information gain (MI) and a Pearson (Person) correlation coefficient, so as to obtain 4 groups of characteristic scores;

The calculation formula of the fischer score method (FS):

wherein x is ₊ 、x _- Respectively positive and negative samples, l ₊ 、l _- Respectively representing the number of positive samples and negative samples;

distance assessment technique (DET):

euclidean distance

Wherein q is _i Representing the average distance, p, of the optimal feature subset _i Representing the distance of each feature of the optimal feature subset;

the information gain (MI) is based on the information measure to evaluate the feature, is the measure of the difference between the prior uncertainty and the expected posterior uncertainty of the feature, and the calculation formula is as follows:

I(x；y)＝H(y)-H(y|x)；

wherein H (y) represents information entropy, and H (y|x) represents conditional entropy; giving a feature x and a label y corresponding to the feature x;

pearson (Person) correlation coefficient measures the correlation between features and classes, and the formula is calculated:

wherein x and y are respectively specific to a given feature and its corresponding label;

step 5.2: sorting the features according to the feature scores obtained in the step 5.1, selecting the top n feature input Radial Basis (RBF) networks with the highest rank to classify, then taking the recognition error rate as a weight value and taking the product of the recognition error rate and the rank of each feature as a new score of the corresponding feature, repeating the operation on 4 groups of feature scores to obtain 4 new scores for each feature, summing the new scores as a weighted score result of the feature, wherein a weighting mechanism is shown in the following formula, and finally reordering according to the sequence from small to large;

S _NEW ＝R _FS E _FS +R _DET E _DET +R _MI E _MI +R _PCC E _PCC ；

Wherein S is _NEW A weighted score representing a feature, R _FS And E is _FS Respectively represent the ranking and the recognition error rate of the characteristic of the Fisher Score (FS) model in the evaluation result of the corresponding model, R _DET And E is _DET Respectively represent the ranking and recognition error rate of the feature in the evaluation result of the corresponding model by the Distance Evaluation Technology (DET), R _MI And E is _MI Respectively representing the ranking and recognition error rate of the characteristic of the information gain (MI) in the evaluation result of the corresponding model, R _PCC And E is _PCC Respectively representing the ranking and the recognition error rate of the feature in the evaluation result of the corresponding model;

step 5.3: the redundant and irrelevant features after ranking are quickly screened out by a Binary Search (BS), the optimal subset is roughly determined, then the subset obtained by the Binary Search (BS) is used as a starting point, the optimal subset is selected by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS) respectively, and finally the two subsets are compared, and the one with the highest recognition accuracy is used as the optimal subset.

Step 6, index standardization: performing index standardization processing on the data in the optimal subset of each type of speed reducer obtained in the step 5; the method comprises the following steps:

the method for standardizing the index in the step 6 comprises the following steps:

the indexes in the index system are classified into larger and more optimal indexes and smaller and more optimal indexes;

The larger and more optimal index standardized expression is:

wherein: x is the index value after standardization; x is an index actual running value; xmax is an index running upper limit; xmin is an index operation lower limit value;

the smaller and more optimal index standardized expression is:

wherein: x is the index value after standardization; x is an index actual running value; x' max is an index operation upper limit value; x' min is the index operation lower limit value.

Step 7, determining subjective weight and objective weight: respectively determining subjective weight and objective weight of the characteristic data indexes after the standardization in the step 6; the method comprises the following steps:

referring to fig. 2, subjective weights are determined: the first step: and constructing a hierarchical structure of the index system. Generally, the three layers are divided into a target layer, a criterion layer and an execution layer. And a second step of: and constructing a judging matrix. The matrix judgment can be obtained through expert discussion, and the relative importance degree of each problem can be determined by using a 'Delphi method'. And a third step of: calculating the relative weight of each layer, and fourth step: consistency test is carried out on the obtained results; the method comprises the following steps:

and 7.1.1, forming a judgment matrix according to the mutual correlation influence between indexes by using an analytic hierarchy process, namely, the importance degree comparison between indexes, wherein the judgment matrix A is defined as follows:

Wherein A is a judgment matrix, and a is an element in the judgment matrix;

the judgment standard quantification of the judgment matrix elements adopts a 1-9 calibration method. In the traditional analytic hierarchy process, because the comparison result has subjectivity, whether the matrix has consistency or not needs to be judged through consistency test, otherwise, the matrix needs to be adjusted, in order to avoid multiple adjustments, an improved analytic hierarchy process is adopted, and the transmission matrix to be optimized for judging the matrix is calculated so as to meet the consistency test. The calculation process is as follows:

step 7.1.2, calculating a to-be-optimized transfer matrix e of the judgment matrix A so as to enable the to-be-optimized transfer matrix e to meet consistency test; converting the judgment matrix A into an antisymmetric matrix B, specifically:

B＝lgA；

wherein A is a judgment matrix, and B is an antisymmetric matrix;

step 7.1.3, obtaining an optimal transfer matrix C by means of solution, wherein the optimal transfer matrix C is specifically:

wherein C is an optimal transfer matrix, and b is an element in an antisymmetric matrix;

step 7.1.4, calculating a to-be-optimized transfer matrix e, which specifically comprises the following steps:

wherein e is a transmission matrix to be optimized, and C is an optimal transmission matrix;

after the transmission matrix to be optimized is obtained, a standard column average method is adopted to obtain a feature vector W corresponding to the maximum feature value, and the vector is a weight vector, and the calculation process is as follows:

Step 7.1.5, according to the to-be-optimized transfer matrix e obtained in step 7.1.4, adopting a canonical column average method to obtain a feature vector W corresponding to the maximum feature value, namely a weight vector; the normalization processing for each column of elements is specifically as follows:

in the method, in the process of the invention,e is a transmission matrix to be optimized;

step 7.1.6, normalized judgment matrixAdding according to columns, specifically:

in the method, in the process of the invention,is a normalized judgment matrix, +.>Is a judgment matrix added according to columns;

step 7.1.7, for the judgment matrix added by columnNormalization processing is carried out to obtain subjective weight w of the characteristic data index _i The method specifically comprises the following steps: />

Wherein i=1, 2, 3..n, w _i Subjective weight w for characteristic data index _i ；

Determining objective weights: the entropy weighting method is characterized in that firstly, feature scaling is carried out on each feature data, and the purpose is as follows: converting the data into dimensionless data, and then solving entropy corresponding to each feature, wherein the purpose is as follows: weighting, and finally, weighting corresponding to each feature; the method comprises the following steps:

the entropy weight method is used for determining the objective weight of the index, and the smaller the abnormality degree of the index is, the smaller the reflected information quantity is, and the lower the corresponding weight value is.

The calculation steps are as follows:

step 7.2.1, performing standardized processing on the evaluation index data to obtain a dimensionless matrix R, wherein the dimensionless matrix R is specifically:

R＝(r _ij ) _m×n ；

Wherein R is a dimensionless matrix, R is an element in the dimensionless matrix R, i is an ith index, and j is a jth index;

step 7.2.2, calculating the information entropy e of the ith index _i The method specifically comprises the following steps:

in the method, in the process of the invention,ei is the information entropy of the ith index, i is the ith index, j is the jth index, and R is an element in the dimensionless matrix R;

step 7.2.3, calculating the entropy vector of the ith index to obtain the objective weight v of the characteristic data index _i The method specifically comprises the following steps:

in the formula, v _i Objective weight of characteristic data index e _i The information entropy of the ith index, i being the ith index.

Step 8, determining the weight value: carrying out combined weighting on the subjective weight and the objective weight in the step 7, and determining a weight; the method comprises the following steps:

subjective weight w using the characteristic data index obtained in step 7.1.7 _i And objective weight v of characteristic data index obtained in step 7.2.3 _i Combining with a combination weighting method to obtain comprehensive weights to obtain an index combination weight vector W _i The method comprises the following steps:

wherein w is _i Subjective weight of characteristic data index, v _i Objective weight of characteristic data index, W _i The weight vectors are combined for the indicators.

Step 9, health status assessment: the health state of the speed reducer is divided by using a gray clustering method and combining the weight value determined in the step 8, see fig. 3, specifically:

Step 9.1, gray clustering method is used for quantitative evaluationThe method for estimating whether the object belongs to a certain gray class is characterized in that the whitening weight of each clustered object for different indexes is calculated through a whitening weight function, so that the gray class is distinguished. Setting n evaluation systems and m evaluation indexes of an evaluation object, and dividing the evaluation object into s different gray classes; the sample value of index j (j= … m) of system i (i=1, … n) is x _ij Dividing the system i into ash classes k (k=1, …, S), wherein the value range of the ash class k is as followsThen->And->Respectively a whitening weight function and a weight of the index j belonging to the gray class k;

step 9.2, calculating the membership degree corresponding to the actual value of the index, wherein the function expression is as follows:

in the method, in the process of the invention,a whitening weight function with index j belonging to gray class k, wherein k is gray class;

step 9.3, calculating the clustering coefficient of the index j on the ash class k according to the step 9.2The method comprises the following steps:

in the method, in the process of the invention,a whitening weight function with index j belonging to gray class k, k being gray class，/>Is a clustering coefficient;

step 9.4, judging what ash class the system i belongs to under the index j, namely what health state the speed reducer is in, dividing according to a whitening weight function, and if the index j is in the range of (1.0-0.8), judging that the speed reducer is in a health state; if the index j is (0.8-0.6), the speed reducer is in a sub-health state; if the index j is within the range of (0.6-0.4), the speed reducer is in a fault state, and if the index j is within the range of (0.4-0), the speed reducer is in a scrapped state.

Speed reducer health state evaluation system based on combination weighting method includes:

the signal acquisition module: randomly collecting acceleration signals, rotation speed signals and current signals of speed reducers with different health states and different types, wherein the module corresponds to the step 1;

a signal data processing module: on embedded edge hardware, noise reduction processing and characteristic value extraction are carried out on acceleration signals and rotation speed signals acquired by a signal acquisition module, a characteristic data set is obtained, the characteristic data in the characteristic data set comprises a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic, and the module corresponds to the step 2;

and a wireless transmission module: transmitting the characteristic data set obtained by the signal data processing module to a remote server database, wherein the module corresponds to the step 3;

the classification processing module: the remote server classifies the speed reducers with different types and different working conditions according to the speed reducer types, the rotating speed signals and the current signals in the signal acquisition module to obtain N types, wherein each type comprises a plurality of speed reducers of the type, and the module corresponds to the step 4;

and an optimal characteristic component extraction module: collecting normal sample data and fault sample data of each type of speed reducer in a classification processing module respectively, selecting optimal characteristic components according to a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic value extracted by a signal data processing module, sequencing the extracted characteristics by using a filtering algorithm, roughly determining an optimal subset by using a Binary Search (BS) and a Radial Basis (RBF) network, selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS), and finally comparing the two subsets to obtain the optimal subset of each type by using the optimal subset with highest recognition accuracy, wherein the module corresponds to the step 5;

Index standardization module: performing index standardization processing on data in the optimal subset of each type of speed reducer obtained by the optimal characteristic component extraction module, wherein the module corresponds to the step 6;

and the subjective weight and objective weight determining module is used for: respectively determining subjective weight and objective weight of the characteristic data index standardized by an index standardization module, wherein the module corresponds to the step 7;

weight determining module: carrying out combined weighting on the subjective weight and the objective weight of the subjective weight and objective weight determining module, and determining a weight value, wherein the module corresponds to the step 8;

health status assessment module: and (3) dividing the health state of the speed reducer by using a gray clustering method and combining the weight determined by the weight determining module, wherein the module corresponds to the step (9).

Speed reducer health state evaluation equipment based on combination weighting method includes:

a memory: a computer program for storing the speed reducer health state assessment method based on the combined weighting method;

a processor: the method is used for realizing the speed reducer health state evaluation method based on the combined weighting method when the computer program is executed.

A computer readable storage medium storing a computer program which, when executed by a processor, enables implementation of a method of assessing a health state of a speed reducer based on a combined weighting method.

Claims (10)

1. The speed reducer health state evaluation method based on the combined weighting method is characterized by comprising the following steps of:

step 1, signal acquisition: randomly collecting acceleration signals, rotation speed signals and current signals of speed reducers with different health states and different types;

step 2, signal data processing: on embedded edge hardware, carrying out noise reduction processing and characteristic value extraction on the acceleration signal and the rotating speed signal acquired in the step 1 to obtain a characteristic data set, wherein the characteristic data in the characteristic data set comprises a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic;

step 3, wireless transmission: transmitting the characteristic data set obtained in the step 2 to a remote server database;

step 4, classification processing: the remote server classifies the speed reducers with different types and different working conditions according to the speed reducer types, the rotating speed signals and the current signals in the step 1 to obtain N types, wherein each type comprises a plurality of speed reducers of the type;

step 5, extracting optimal characteristic components: collecting normal sample data and fault sample data of each type of speed reducer in the step 4 respectively, selecting optimal characteristic components according to the time domain characteristic values, the frequency domain characteristic values and the characteristic value domain characteristic values extracted in the step 2, sequencing the extracted characteristics by using a filtering algorithm, determining an optimal subset by using a Binary Search (BS) and a Radial Basis (RBF) network, selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS), and finally comparing the two subsets to obtain the optimal subset with the highest recognition accuracy among the two subsets;

Step 6, index standardization: performing index standardization processing on the data in the optimal subset of each type of speed reducer obtained in the step 5;

step 7, determining subjective weight and objective weight: respectively determining subjective weight and objective weight of the characteristic data indexes after the standardization in the step 6;

step 8, determining the weight value: carrying out combined weighting on the subjective weight and the objective weight in the step 7, and determining a weight;

step 9, health status assessment: and (3) dividing the health state of the speed reducer by using a gray clustering method and combining the weight determined in the step (8).

2. The method for estimating health of a speed reducer based on a combined weighting method according to claim 1, wherein in the step 1, an edge node is installed on a bearing seat of an input shaft and a bearing seat of an output shaft of the speed reducer, and a high-precision acceleration sensor is integrated inside the edge node to collect acceleration signals; acquiring a rotating speed signal by externally connecting a photoelectric tachometer on an edge node of an input shaft bearing seat of a speed reducer; collecting current signals by externally connecting a current sensor on an independent edge node; the model of the high-precision acceleration sensor is 608A11, and the frequency range is 0.5000-10000Hz.

3. The method for estimating health state of a speed reducer based on a combined weighting method according to claim 1, wherein the extracting of the optimal feature component in the step 5 is specifically:

step 5.1: collecting characteristic data sets of each type of speed reducer under normal operation and under different fault conditions, and evaluating fault characteristics from distance measurement, information measurement and correlation measurement by using 4 filtering algorithms, namely a Fisher score method (FS), a Distance Evaluation Technology (DET), an information gain (MI) and a Pearson (Person) correlation coefficient, so as to obtain 4 groups of characteristic scores;

the calculation formula of the fischer score method (FS):

wherein x is ₊ 、x _- Respectively positive and negative samples, l ₊ 、l _- Respectively representing the number of positive samples and negative samples;

distance assessment technique (DET):

euclidean distance

Wherein q is _i Representing the average distance, p, of the optimal feature subset _i Representing the distance of each feature of the optimal feature subset;

the information gain (MI) is based on the information measure to evaluate the feature, is the measure of the difference between the prior uncertainty and the expected posterior uncertainty of the feature, and the calculation formula is as follows:

I(x；y)＝H(y)-H(y|x)；

wherein H (y) represents information entropy, and H (y|x) represents conditional entropy; giving a feature x and a label y corresponding to the feature x;

Pearson (Person) correlation coefficient measures the correlation between features and classes, and the formula is calculated:

wherein x and y are respectively specific to a given feature and its corresponding label;

step 5.2: sorting the features according to the feature scores obtained in the step 5.1, selecting the top n feature input Radial Basis (RBF) networks with the highest rank to classify, then taking the recognition error rate as a weight value and taking the product of the recognition error rate and the rank of each feature as a new score of the corresponding feature, repeating the operation on 4 groups of feature scores to obtain 4 new scores for each feature, summing the new scores as a weighted score result of the feature, wherein a weighting mechanism is shown in the following formula, and finally reordering according to the sequence from small to large;

S _NEW ＝R _FS E _FS +R _DET E _DET +R _MI E _MI +R _PCC E _PCC ；

wherein S is _NEW A weighted score representing a feature, R _FS And E is _FS Respectively represent the ranking and the recognition error rate of the characteristic of the Fisher Score (FS) model in the evaluation result of the corresponding model, R _DET And E is _DET Respectively represent the ranking and recognition error rate of the feature in the evaluation result of the corresponding model by the Distance Evaluation Technology (DET), R _MI And E is _MI Respectively representing the ranking and recognition error rate of the characteristic of the information gain (MI) in the evaluation result of the corresponding model, R _PCC And E is _PCC Respectively represent the rows of the features in the evaluation results of the corresponding models Name and recognition error rate;

step 5.3: and (3) rapidly screening redundant and irrelevant features after ranking by using a Binary Search (BS), determining an optimal subset, then selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS) respectively with the subset obtained by the Binary Search (BS) as a starting point, and finally comparing the two subsets to obtain the optimal subset with the one with the highest recognition accuracy.

4. The method for estimating health state of a speed reducer based on a combined weighting method according to claim 1, wherein the method for normalizing the index in step 6 is as follows:

the indexes in the index system are classified into larger and more optimal indexes and smaller and more optimal indexes;

the larger and more optimal index standardized expression is:

wherein: x is the index value after standardization; x is an index actual running value; xmax is an index running upper limit; xmin is an index operation lower limit value;

the smaller and more optimal index standardized expression is:

wherein: x is the index value after standardization; x is an index actual running value; x' max is an index operation upper limit value; x' min is the index operation lower limit value.

5. The method for estimating health state of a speed reducer based on a combined weighting method according to claim 1, wherein the determining subjective weight in the step 7 is specifically:

And 7.1.1, comparing the two indexes by two according to the mutual correlation influence by using an analytic hierarchy process to form a judgment matrix, wherein the judgment matrix A is defined as follows:

wherein A is a judgment matrix, and a is an element in the judgment matrix;

step 7.1.2, calculating a to-be-optimized transfer matrix e of the judgment matrix A so as to enable the to-be-optimized transfer matrix e to meet consistency test; converting the judgment matrix A into an antisymmetric matrix B, specifically:

B＝lgA；

wherein A is a judgment matrix, and B is an antisymmetric matrix;

step 7.1.3, obtaining an optimal transfer matrix C by means of solution, wherein the optimal transfer matrix C is specifically:

wherein C is an optimal transfer matrix, and b is an element in an antisymmetric matrix;

step 7.1.4, calculating a to-be-optimized transfer matrix e, which specifically comprises the following steps:

wherein e is a transmission matrix to be optimized, and C is an optimal transmission matrix;

step 7.1.5, according to the to-be-optimized transfer matrix e obtained in step 7.1.4, adopting a canonical column average method to obtain a feature vector W corresponding to the maximum feature value, namely a weight vector; the normalization processing for each column of elements is specifically as follows:

in the method, in the process of the invention,e is a transmission matrix to be optimized;

step 7.1.6, adding the normalized judgment matrix m according to columns, specifically:

in the method, in the process of the invention,is a normalized judgment matrix, +. >Is a judgment matrix added according to columns;

step 7.1.7, for the judgment matrix added by columnNormalization processing is carried out to obtain subjective weight wi of the characteristic data index, and the subjective weight wi is specifically as follows:

wherein i=1, 2, 3..n, w _i Subjective weight w for characteristic data index _i ；

The objective weight determination in the step 7 is specifically:

step 7.2.1, performing standardized processing on the evaluation index data to obtain a dimensionless matrix R, wherein the dimensionless matrix R is specifically:

R＝(r _ij ) _m×n ；

wherein R is a dimensionless matrix, R is an element in the dimensionless matrix R, i is an ith index, and j is a jth index;

step 7.2.2, calculating the information entropy e of the ith index _i The method specifically comprises the following steps:

in the method, in the process of the invention,e _i information entropy of the ith index, i is the ith index, j is the jth index, and R is an element in the dimensionless matrix R;

step 7.2.3, calculating the entropy vector of the ith index to obtain the objective weight v of the characteristic data index _i The method specifically comprises the following steps:

in the formula, v _i Objective weight of characteristic data index e _i The information entropy of the ith index, i being the ith index.

6. The method for estimating health state of a speed reducer based on a combined weighting method according to claim 1, wherein the determining weight in the step 8 is specifically:

subjective weight w using the characteristic data index obtained in step 7.1.7 _i And objective weight v of characteristic data index obtained in step 7.2.3 _i Combining with a combination weighting method to obtain comprehensive weights to obtain an index combination weight vector W _i The method comprises the following steps:

wherein w is _i Subjective weight of characteristic data index, v _i Objective weight of characteristic data index, W _i The weight vectors are combined for the indicators.

7. The method for estimating health state of a speed reducer based on the combined weighting method according to claim 1, wherein the estimating health state in step 9 is specifically:

step 9.1, setting n evaluation systems and m evaluation indexes of an evaluation object, wherein the n evaluation systems and the m evaluation indexes are divided into s different gray classes; the index j (j=1, … m) of the system i (i=1, … n) has a sample value x _ij Dividing the system i into ash classes k (k=1, …, s), wherein the value range of the ash class k is as followsThen->And->Respectively a whitening weight function and a weight of the index j belonging to the gray class k;

step 9.2, calculating the membership degree corresponding to the actual value of the index, wherein the function expression is as follows:

in the method, in the process of the invention,a whitening weight function with index j belonging to gray class k, wherein k is gray class;

step 9.3, calculating the clustering coefficient of the index j on the ash class k according to the step 9.2The method comprises the following steps:

in the method, in the process of the invention,white with index j belonging to gray class kThe weight function, k is gray class, +. >Is a clustering coefficient;

step 9.4, judging what ash class the system i belongs to under the index j, namely what health state the speed reducer is in, dividing according to a whitening weight function, and if the index j is in the range of (1.0-0.8), judging that the speed reducer is in a health state; if the index j is (0.8-0.6), the speed reducer is in a sub-health state; if the index j is within the range of (0.6-0.4), the speed reducer is in a fault state, and if the index j is within the range of (0.4-0), the speed reducer is in a scrapped state.

8. Speed reducer health state evaluation system based on combination weighting method, characterized by comprising:

the signal acquisition module: randomly collecting acceleration signals, rotation speed signals and current signals of speed reducers with different health states and different types;

a signal data processing module: on embedded edge hardware, carrying out noise reduction processing and characteristic value extraction on the acceleration signal and the rotating speed signal acquired by the signal acquisition module to obtain a characteristic data set, wherein the characteristic data in the characteristic data set comprises a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic;

and a wireless transmission module: transmitting the characteristic data set obtained by the signal data processing module to a remote server database;

The classification processing module: the remote server classifies the speed reducers with different types and different working conditions according to the speed reducer types, the rotating speed signals and the current signals in the signal acquisition module to obtain N types, wherein each type comprises a plurality of speed reducers of the type;

and an optimal characteristic component extraction module: the method comprises the steps of respectively collecting normal sample data and fault sample data of each type of speed reducer in a classification processing module, selecting optimal characteristic components according to a time domain characteristic value, a frequency domain characteristic value and a characteristic value domain characteristic value extracted by a signal data processing module, sequencing the extracted characteristics by using a filtering algorithm, determining an optimal subset by using a Binary Search (BS) and a Radial Basis (RBF) network, selecting the optimal subset by using a Sequence Forward Search (SFS) and a Sequence Backward Search (SBS), and finally comparing the two subsets to obtain the optimal subset of each type, wherein the optimal subset has the highest recognition accuracy;

index standardization module: performing index standardization processing on the data in the optimal subset of each type of speed reducer obtained by the optimal characteristic component extraction module;

and the subjective weight and objective weight determining module is used for: respectively determining subjective weight and objective weight of the characteristic data index standardized by the index standardization module;

Weight determining module: the subjective weight and the objective weight of the subjective weight and objective weight determining module are combined and weighted to determine a weight value;

health status assessment module: and dividing the health state of the speed reducer by using a gray clustering method and combining the weight determined by the weight determining module.

9. Speed reducer health state assessment equipment based on combination weighting method, characterized by comprising:

a memory: a computer program for storing the speed reducer health state assessment method based on the combined weighting method;

a processor: the method is used for realizing the speed reducer health state evaluation method based on the combined weighting method when the computer program is executed.

10. A computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program when executed by a processor can implement a method for estimating health state of a speed reducer based on a combined weighting method.