CN110245379A - A Method for Identifying the Failure Mechanism of a Sealed Electromagnetic Relay - Google Patents

Abstract

The invention relates to a method for discriminating the failure mechanism of a sealed electromagnetically driven switching device. The method is simple and easy to implement, has a wide range of applications, low cost, high accuracy, can remove interference factors such as arc erosion, does not need to open the cavity of the tested relay, and does not need to use a high-power microscope. The specific steps are as follows: 1. Collect the life of the relay The six key performance degradation parameter data in the period, the collected data matrix is used as the data set for the identification of the failure mechanism of the electromagnetic relay. 2. Use the FIR (Finite Impulse Response, FIR) high-pass filter to filter the data set obtained from the test to remove the interference caused by the randomness of arc erosion and material transfer to the data acquisition system. 3. Using the Bayesian discriminant method to extract the best combination parameters that can distinguish the failure mechanism of the relay from the data set after the noise reduction filter. 4. Use the random forest algorithm to discriminate the relay failure mechanism through the optimal combination of parameters.

Description

A Method for Identifying the Failure Mechanism of a Sealed Electromagnetic Relay

Technical field:

The invention relates to a method for discriminating the failure mechanism of a sealed electromagnetic relay.

Background technique:

The contact failure of electromagnetic relay is mainly caused by the arcing of its contacts during the pull-in and release process. At present, the methods for identifying the failure mechanism of relays include post-observation method, monitoring dynamic contact resistance method and principal component analysis method. The after-the-fact observation method is usually to conduct a complete life test on the tested relay, open the case after it fails, and use high-powered microscopes and other equipment to observe the distribution of contact surface materials to determine the failure mechanism. This method has high accuracy, high cost, narrow application range, and sometimes cannot be completed due to the limitation of objective conditions; although the monitoring dynamic contact resistance method realizes the online identification of relay failure mechanism, it only analyzes a single performance degradation parameter and ignores other affecting relays. The performance degradation parameters of the failure mechanism, so its discrimination accuracy and versatility are insufficient; the principal component analysis method is a data feature extraction method based on linear transformation of the original feature data, and the relay performance degradation parameter data is nonlinear data. Meta-analysis will ignore these non-linear attributes after transformation, which will have a certain impact on the accuracy of discrimination.

Invention content:

The purpose of the present invention is to provide a method for identifying the failure mechanism of a sealed electromagnetic relay. The above-mentioned purpose is achieved through the following specific steps:

Step 1: Use the developed relay reliability life test system to collect data on six key performance degradation parameters in the typical relay life cycle: overtravel time, bounce time, contact resistance, pull-in time, release time and arcing time. The collected matrix data Data _n×6 is used as the data set for the identification of the contact failure mechanism of the electromagnetic relay;

Step 2: Considering that the randomness of the ignition arc and material transfer will bring certain interference to the data acquisition system, the data set obtained from the test is processed by FIR (Finite Impulse Response, FIR) high-pass filter for noise reduction filtering;

Step 3: using the Bayesian discriminant method to extract the optimal discriminant combination parameters that can distinguish the failure mechanism of the relay from the data set processed by the noise reduction filter. When P(C _i |X)>P(C _j |X)(1≤i,j≤m,j≠i) is satisfied, the samples to be classified in the data set after noise reduction processing X={a ₁ ,a ₂ ,...,a _n }, divided into categories C _i (1≤i≤m), it can be known from Bayes theorem:

Among them, P(C _i )=P(C _j ),(C _i ,C _j ,i≠j), that is, the probability of each category in the data set is equal, and the maximization of P(C _j |X) is:

Assuming that the conditional attributes in the Bayesian algorithm are independent of each other, there are:

in, S _i represents the number of instances of C _i samples in the training set, S represents the total number of samples in the training set, then the model can be expressed as:

Probability P(a ₁ |C _i ),P(a ₂ |C _i ),...,P(a _n |C _i ), can be estimated by training samples, where:

in, represents the Gaussian density function of attribute A _k , represent the standard deviation and mean, respectively.

For the sample X whose failure mechanism is to be classified, the conditional probability P(C _i )P(X|C _i ) of each category C _i ∈ C is calculated. When P(C _i |X)>P(C _j |X)(1≤i, j≤m, j≠i) is satisfied, the sample X to be classified is assigned to the C _i category. The discrimination ability of each degradation parameter to the contact failure mechanism is shown in Table 1. The larger the discrimination ability value is, the stronger the discrimination ability of the corresponding degradation parameter is to the contact failure mechanism. Therefore, the best discriminative combination parameters to distinguish the relay failure mechanism are overtravel time and bounce time.

Table 1 Discriminant ability value of each degradation parameter for contact failure mechanism

Step 4: Use the random forest algorithm to discriminate the contact failure mechanism of the relay with the combined parameters of overtravel time and bounce time. Use the sampling method with replacement to sample the samples in the data set processed by the noise reduction filter, and then construct a suitable decision tree for each sampled sample to form a random forest model, and vote for the test sample with each constructed decision tree, The category with the most votes is the category to which the test sample belongs. Random Forest uses a marginal function to measure the degree to which the average number of correct classifications in the model exceeds the average number of misclassifications. The larger the value of the marginal function, the better the classification effect. The definition of the marginal function is shown in formula (6).

MR(x,y)=I _θ (f(x,θ)=y)-max _j=y I _θ (f(x,θ)=j) (6)

The generalization error of random forest model in two-dimensional space can be expressed by formula (7).

PE=p _x,y (MR(x,y)<0) (7)

When there are enough decision trees, the random forest classification model obeys the law of large numbers, and the convergence of random variables in the model can be expressed by formula (8). Equation (8) shows that the expansion of random forest is very good, and there will be no overfitting phenomenon with the expansion of subtrees.

Beneficial effects of the present invention:

The invention discloses a method for discriminating the failure mechanism of a sealed electromagnetic relay. The developed relay reliability life test system is used to discriminate the failure mechanism (bridging failure, wear failure, pollution failure) of the collected test data, without the need for testing the relay. The cavity opening treatment is performed without the use of a high-power microscope. This method is simple and easy to implement, has a wide range of applications, low cost, high accuracy, and can remove interference factors such as arc erosion.

Description of drawings:

Fig. 1 is a hardware system block diagram of the present invention.

Fig. 2 is a block diagram of the software system of the present invention.

Fig. 3 is a schematic diagram of the control circuit of the present invention.

Fig. 4 is a schematic diagram of the contact monitoring circuit of the present invention.

Fig. 5 is a schematic diagram of the degradation data acquisition circuit of the present invention.

Fig. 6 is a schematic diagram of the power supply circuit of the present invention.

Fig. 7 is a schematic diagram of the surface morphology of the contacts under the normal state and three different failure mechanism states of the present invention.

Fig. 8 is a diagram of the discrimination accuracy rate of different subtrees of the random forest of the present invention.

Fig. 9 is a discrimination effect diagram of random forest 5 subtrees of the present invention.

Fig. 10 is a discrimination effect diagram of random forest 10 subtrees of the present invention.

Fig. 11 is a discrimination effect diagram of the random forest 20 subtrees of the present invention.

Fig. 12 is a discrimination effect diagram of the random forest 50 subtrees of the present invention.

Detailed ways:

Example 1:

The invention relates to a method for discriminating the failure mechanism of a sealed electromagnetic relay. The above-mentioned purpose is achieved through the following specific steps:

Step 1: Use the developed relay reliability life test system to collect data on six key performance degradation parameters in the typical relay life cycle: overtravel time, bounce time, contact resistance, pull-in time, release time and arcing time. The collected matrix data Data _n×6 is used as the data set for the identification of the contact failure mechanism of the electromagnetic relay;

Step 2: Considering that the randomness of the ignition arc and material transfer will bring certain interference to the data acquisition system, the data set obtained from the test is processed by FIR (Finite Impulse Response, FIR) high-pass filter for noise reduction filtering;

Step 3: using the Bayesian discriminant method to extract the optimal discriminant combination parameters that can distinguish the failure mechanism of the relay from the data set processed by the noise reduction filter. When P(C _i |X)>P(C _j |X)(1≤i,j≤m,j≠i) is satisfied, the samples to be classified in the data set after noise reduction processing X={a ₁ ,a ₂ ,...,a _n }, divided into categories C _i (1≤i≤m), it can be known from Bayes theorem:

Among them, P(C _i )=P(C _j ),(C _i ,C _j ,i≠j), that is, the probability of each category in the data set is equal, and the maximization of P(C _j |X) is:

Assuming that the conditional attributes in the Bayesian algorithm are independent of each other, there are:

in, S _i represents the number of instances of C _i samples in the training set, S represents the total number of samples in the training set, then the model can be expressed as:

Probability P(a ₁ |C _i ),P(a ₂ |C _i ),...,P(a _n |C _i ), can be estimated by training samples, where:

in, represents the Gaussian density function of attribute A _k , represent the standard deviation and mean, respectively.

For the sample X whose failure mechanism is to be classified, the conditional probability P(C _i )P(X|C _i ) of each category C _i ∈ C is calculated. When P(C _i |X)>P(C _j |X)(1≤i, j≤m, j≠i) is satisfied, the sample X to be classified is assigned to the C _i category. The discrimination ability of each degradation parameter to the contact failure mechanism is shown in Table 1. The larger the discrimination ability value is, the stronger the discrimination ability of the corresponding degradation parameter is to the contact failure mechanism. Therefore, the best discriminative combination parameters to distinguish the relay failure mechanism are overtravel time and bounce time.

Table 1 Discriminant ability value of each degradation parameter for contact failure mechanism

Step 4: Use the random forest algorithm to discriminate the contact failure mechanism of the relay with the combined parameters of overtravel time and bounce time. Use the sampling method with replacement to sample the samples in the data set processed by the noise reduction filter, and then construct a suitable decision tree for each sampled sample to form a random forest model, and vote for the test sample with each constructed decision tree, The category with the most votes is the category to which the test sample belongs. Random Forest uses a marginal function to measure the degree to which the average number of correct classifications in the model exceeds the average number of misclassifications. The larger the value of the marginal function, the better the classification effect. The definition of the marginal function is shown in formula (6).

MR(x,y)=I _θ (f(x,θ)=y)-max _j=y I _θ (f(x,θ)=j) (6)

The generalization error of random forest model in two-dimensional space can be expressed by formula (7).

PE=p _x,y (MR(x,y)<0) (7)

When there are enough decision trees, the random forest classification model obeys the law of large numbers, and the convergence of random variables in the model can be expressed by formula (8). Equation (8) shows that the expansion of random forest is very good, and there will be no overfitting phenomenon with the expansion of subtrees.

Example 2:

A method for identifying the failure mechanism of a sealed electromagnetic relay. Using the distributed control method, the data of six key performance degradation parameters, including overtravel time, bounce time, contact resistance, pull-in time, release time and arcing time, are monitored synchronously in the entire life cycle of 8 tested relays for collection and storage. The hardware system includes a control circuit, a relay contact monitoring circuit, a degradation data acquisition circuit and a power supply circuit. The block diagram of the hardware system is shown in Figure 1. The software system adopts the data acquisition control DAQ developed by Advantech, based on the Windows XP system environment, written with VB6.0 software. Software system block diagram shown in Figure 2 . When the contact failure of the tested relay occurs, the system can stop immediately and record the current number of actions and status of the contact, and send out an alarm at the same time. The system is designed according to GB/T 15510-2008 "General Rules for Reliability Test of Electromagnetic Relays for Control" and GB2423 "Basic Environmental Test Regulations for Electrical and Electronic Products". The specific technical indicators are as follows:

(1) Coil excitation voltage: the range is DC 0-100V, ±10% of the rated voltage;

(2) Load power supply: ±10% of rated voltage, the upper limit is DC 80A/100V;

(3) Number of monitoring relays: 8;

(4) Detection parameters: overtravel time, bounce time, contact resistance, pull-in time, release time, arcing time;

(5) Measurement accuracy: 1μs;

(6) Action frequency: optional within the range of 30 to 600 times/min;

(7) Monitoring start and end time: the time period after 40% of the time after the coil of the tested relay is powered on and after 40% of the time after its coil is powered off;

(8) Data storage: The data can be manually saved in real time during the test, and can be automatically saved at regular intervals.

Example 3:

Control circuit design:

Using the IPC-610L Advantech industrial computer as the upper computer, the Modbus communication protocol is used to issue high and low level commands to the Y0~Y7 coil pins of the lower computer controller (DVP-32EH Delta PLC) to complete the pull-in and contact of the relay contacts. separate. Use Yutai UT-891-USB to 485 serial port communication cable to centrally control the two PLCs. The schematic diagram is shown in Figure 3.

Contact monitoring circuit design:

Realize synchronous monitoring and storage of six key performance return parameter data of overtravel time, bounce time, contact resistance, pull-in time, release time, and arcing time.

Use 1Ω and 0.1Ω/50W sampling resistors to convert the contact current and coil current to voltage. When contact failure occurs in the contacts of the relay under test, the load power supply is cut off by opening the switch K1. The schematic diagram is shown in Figure 4.

Degradation data acquisition circuit design:

Receive the signal data monitored by the contact monitoring circuit through the Advantech ADAM-3968 adapter board, and transmit the received data to the Advantech PCI-1747U high-performance data acquisition card through the Advantech PCL-10168 communication cable. The schematic diagram is shown in Figure 5 shown.

Power supply circuit design:

The LRS-350-5 and LRS-350-12 switching power supplies produced by Taiwan Mean Well Company are used to supply power to the public pins (C0~C3) of the DVP-32EH Delta PLC and the coil of the relay under test respectively. The schematic diagram is shown in the figure 6.

As mentioned above, the high temperature and arcing generated during the pull-in and release of the relay will corrode the contact material, causing the transfer of material between the contacts, resulting in changes in the contact pressure on the contact surface, the contact gap and the contact overtravel , with the gradual accumulation of material transfer, it will lead to different types of contact failures in the contacts. Fig. 7 is a schematic diagram of the surface morphology of the contacts under the normal state of the relay and three different failure mechanism states. Assume that the contact material in each sub-graph is transferred from the lower static contact to the upper movable contact. The normal state is shown in Figure 7(a), the contact pressure on the surface of the contact pair is relatively uniform, and the contact gap and contact overtravel are not beyond the normal range; due to the cumulative effect of high temperature and arcing between the contacts, the It causes the contact material to splash and evaporate within a wide range of the contact spot of the other party, resulting in a decrease in the contact overtravel and an increase in the contact gap. When the contact overtravel reaches the limit value of zero, the contact pair still cannot After the contact is completed, "wear" failure occurs, as shown in Figure 7(b); when the contact material accumulates in a certain area on the contact surface of the other party to form one or more stable protrusions, the contact gap is reduced. When the limit is reached, the contact gap is completely filled, causing the contact pair of the relay to still be in contact when the relay is disconnected, and a "bridging" failure occurs, as shown in Figure 7(c); when the total amount of contact material received and lost , the direction is roughly the same, the contact gap and contact overtravel are basically within the normal range, but the surface of the contact pair is severely ablated, and various pollutants are deposited. When the limit is reached, the static contact is completely "burned through" , a "pollution" failure occurs, as shown in Figure 7(d). The contact failure discrimination effect of the density electromagnetic relay contact of the present invention is shown in Figs. 8-12.

Claims (1)

1. A failure mechanism judging method for a sealed electromagnetic drive switch electric appliance comprises the following specific steps:

the method comprises the following steps: collecting six key performance degradation parameter Data of overtravel time, bounce time, contact resistance, pull-in time, release time and arcing time in a typical relay life cycle by utilizing a developed relay reliability life test system, and collecting matrix Data_n×6The data set is used as a data set for judging the contact failure mechanism of the electromagnetic relay;

step two: considering that contact arcing and randomness of material transfer can bring certain interference to a data acquisition system, an FIR (Finite Impulse Response) high-pass filter is adopted to carry out noise reduction filtering processing on a data set obtained by a test;

step three: and extracting the optimal identification combination parameters capable of distinguishing the failure mechanism of the relay from the data set after noise reduction and filtering by adopting a Bayesian discrimination method. When P (C) is satisfied_i|X)>P(C_jIf | X) (i is not less than 1, j is not more than m, and j is not equal to i), the sample X to be classified in the data set after noise reduction processing is set to { a ═ a₁,a₂,...,a_nIs classified into class C_i(i is more than or equal to 1 and less than or equal to m), and the Bayes theorem shows that:

wherein, P (C)_i)＝P(C_j),(C_i,C_jI ≠ j), namely: the probability of each class in the dataset is equal, pair P (C)_j| X) maximized are:

the Bayesian algorithm is provided with the following conditional attributes which are independent of each other:

wherein,S_iis represented by C_iThe number of instances of the samples in the training set, S representing the total number of samples in the training set, the model can be expressed as:

probability P (a)₁|C_i),P(a₂|C_i),...,P(a_n|C_i) May be estimated from training samples, wherein:

wherein,representative Attribute A_kThe function of the gaussian density of (a),respectively, the standard deviation and the mean thereof.

For the sample X to be classified of the relay failure mechanism, calculating each class C_iC (C) conditional probability_i)P(X|C_i). When P (C) is satisfied_i|X)>P(C_jWhen | X) (i is more than or equal to 1, m is more than or equal to j, and j is not equal to i), the sample X to be classified is divided into C_iA category.

Step four: and judging the contact failure mechanism of the relay by adopting a random forest algorithm and combining parameters of the overtravel time and the bounce time. Sampling samples in the data set after noise reduction and filtering processing by using a replaced sampling method, constructing a proper decision tree for each sampling sample to form a random forest model, voting the test samples by using the constructed decision trees, wherein the category with the most votes is the category to which the test sample belongs. The random forest measures the degree that the average correct classification number exceeds the average wrong classification number in the model by using a marginal function, the larger the value of the marginal function is, the better the classification effect is represented, and the definition of the marginal function is shown in formula (6).

MR(x,y)＝I_θ(f(x,θ)＝y)-max_j＝yI_θ(f(x,θ)＝j) (6)

The generalization error of the random forest model in the two-dimensional space can be represented by formula (7).

PE＝p_x,y(MR(x,y)<0) (7)

When the decision tree is large enough, the random forest classification model obeys the law of large numbers, and the convergence condition of the random variable in the model can be represented by formula (8). Formula (8) shows that the random forest has good expansibility, and no overfitting phenomenon occurs along with the expansion of the subtrees.