A high-precision molecular pump fault diagnosis method and model building method

Abstract

The present invention provides a high-precision molecular pump fault diagnosis method and a model building method thereof. First, a fault diagnosis data set is collected, including the failure state of a ceramic ball bearing, the slight fatigue state of a rubber vibration damping pad on a ceramic ball bearing, the severe fatigue state of a rubber vibration damping pad, and the vibration signal of a normal state; then, based on the method of time domain and frequency domain analysis, a feature extraction data set based on time domain and frequency domain analysis is obtained, and then, through the designed 1DANN neural network diagnosis model, training and deployment are performed to effectively judge the molecular pump fault diagnosis data. The present invention combines the PSO algorithm on the basis of the 1DANN neural network to ensure that it avoids falling into a local optimal solution during the calculation process. At the same time, combined with the five-fold cross-validation algorithm, model overfitting can be avoided, thereby finding the optimal fitness value, obtaining the optimal parameters, and training the optimal neural network model, thereby improving the accuracy of molecular pump fault diagnosis.

Description

Technical Field

[0001] The present invention belongs to the technical field of vacuum molecular pump fault diagnosis in semiconductor processes, and in particular relates to a high-precision molecular pump fault diagnosis method and a model building method thereof. Background Art

[0002] A molecular pump is a high-efficiency vacuum pump used for exhaust in a vacuum environment. It achieves vacuum extraction by transporting gas molecules from an area with higher exhaust gas pressure to an area with lower exhaust pressure through the conversion of molecular motion. In the process flow of semiconductor manufacturing, molecular pumps are a key vacuum pump technology used to create and maintain an ultra-high vacuum (UHV) environment. In semiconductor manufacturing, various process steps need to be performed in a dust-free and impurity-free environment to ensure the quality and performance of the final product. In semiconductor manufacturing, molecular pumps are often used in key process steps such as physical vapor deposition (PVD), chemical vapor deposition (CVD), and ion implantation. These steps need to be performed in a highly clean environment to ensure that the materials deposited on semiconductor devices have consistent performance and quality. Molecular pumps effectively achieve this goal by pumping gas out of the vacuum chamber.

[0003] Sudden failure of molecular pumps may destroy the vacuum environment of the system, leading to unpredictable losses caused by production or experimental failure. In the prior art, for sudden failure of molecular pumps, the faulty molecular pumps are usually maintained or replaced after the failure occurs, which cannot be prevented in advance to avoid accidents. In order to diagnose molecular pump failures in a timely and accurate manner, many special fault diagnosis methods and systems have appeared in the prior art. These methods and systems usually combine sensor technology, data acquisition and processing technology, and artificial intelligence algorithms, etc., which can monitor the operating status of molecular pumps in real time, analyze data, and provide timely alarms and fault diagnosis results when abnormal conditions are found. At present, machine learning diagnostic algorithms have achieved good (or very good) results when the data set is balanced. However, in the face of actual industrial operation conditions, there is very little molecular pump failure data in the process flow of semiconductor manufacturing, and the data set is unbalanced, resulting in low fault diagnosis accuracy of machine learning algorithms and model overfitting problems. Summary of the invention

[0004] In view of the above problems, the first aspect of the present invention provides a method for building a high-precision molecular pump fault diagnosis model, comprising the following process:

[0005] Step 1, collecting vibration signals of the molecular pump under different states, including vibration signals of a ceramic ball bearing failure state, a rubber vibration damping pad on the ceramic ball bearing slightly fatigued state, a rubber vibration damping pad severely fatigued state, and a normal state;

[0006] Step 2, performing time domain and frequency domain analysis on the collected data to obtain a feature-extracted data set based on the time domain and frequency domain analysis;

[0007] Step 3, selecting frequency domain features of different states and combining them with time domain features to construct a vibration feature data set of different states of the molecular pump, and dividing the feature data set into a training set and a test set;

[0008] Step 4, design a 1DANN fault diagnosis model based on swarm intelligence search optimization PSO algorithm and train it using the feature data set;

[0009] Step 5: Use the test set to test the trained 1DANN fault diagnosis model and obtain the final 1DANN fault diagnosis model.

[0010] Preferably, ceramic ball bearings that fail in the full life experiment, rubber vibration damping pads of ceramic ball bearings with slight fatigue, and rubber vibration damping pads of ceramic ball bearings with severe fatigue are collected, and experiments are carried out under four conditions (ceramic ball bearing failure category 1, rubber vibration damping pad slight fatigue category 2, rubber vibration damping pad severe fatigue category 3, and normal operating condition category 4) by replacing parts; the sampling frequency is set to 1kHz, and when collecting experimental data, the peak-to-peak value of the signal generator is set to 15V, the frequency is set to 2.25kHz, the signal is manually triggered, the number of cycles is set to 9000, and 20,000 normal data are collected through multiple triggering; the ceramic ball bearings are replaced, and 8,000 category 1 fault data are collected; the rubber vibration damping pads with slight fatigue are replaced, and 6,000 category 2 fault data are collected; the rubber vibration damping pads with severe fatigue are replaced, and 4,000 category 2 fault data are collected, and finally the molecular pump fault diagnosis data set X is obtained, and the sampling period is 0.001s.

[0011] Preferably, the time domain and frequency domain analysis is performed in step 2, specifically:

[0012] For the collected sample data X, assuming that N represents the number of sampling points collected in a period, x i represents the i-th data sample in the current cycle, then i∈[1,N];

[0013] Select the maximum value X in the time domain analysis max =max(x i ), minimum value X min =min(x i ), peak-to-peak value X ff =X max -X min ,average value variance RMS Crest Factor Pulse Factor As a vibration state analysis, it better reflects the hidden characteristics of data;

[0014] In frequency domain analysis, the time domain signal is converted into a frequency domain signal through the fast Fourier transform FFT. The FFT formula is:

[0015]

[0016] Where k = 0, 1, ..., N-1 represents the signal length, is the rotation factor, f k is the frequency domain value, x(n) is the sampling point in the time domain, n is the sequence index of the time domain sampling point, and k is the index of the frequency domain value;

[0017] Use frequency domain mean Frequency Domain Center of Gravity Frequency domain standard deviation The three-state frequency domain characteristics describe the vibration state information of the molecular pump;

[0018] The data x i The maximum value, minimum value, peak-to-peak value, average value, variance, root mean square, peak factor, pulse factor in the time domain analysis are combined with the frequency domain mean, frequency domain centroid, and frequency domain standard deviation in the frequency domain analysis to form a feature vector f i , the final input sampling data set X becomes the feature extracted data set F based on time domain and frequency domain analysis.

[0019] Preferably, the step 3 is specifically:

[0020] After feature extraction based on time domain and frequency domain analysis, the dataset F is obtained. Every 10 data are sliced ​​into a group of samples. 2000 groups of normal vibration feature data, 800 groups of ceramic ball bearing failure data, 600 groups of slightly fatigued rubber vibration damping pad failure vibration feature data, and 400 groups of severely fatigued rubber vibration damping pad failure vibration feature data are obtained. The training set and the test set are divided into 8:2 to obtain the molecular pump vibration training set F. train With the test set F val .

[0021] Preferably, the 1DANN fault diagnosis model includes an input layer, an output layer and a hidden layer;

[0022] The input layer is mainly composed of a fully connected layer, which is used to input the molecular pump training data set F train The 11-dimensional features are mapped to the 120-dimensional implicit function feature space; the main formula of the input layer is as follows:

[0023]

[0024] in is the feature mapping matrix;

[0025] The hidden layer is specifically, assuming is the hidden layer input variable, y is the output variable, k m For hidden neuron output, there exists a mapping relationship where g is the activation function; assuming v i,j is the weight of the i-th input variable and the j-th hidden layer neuron, is the threshold of the jth neuron in the hidden layer k, i.e., the bias term; we can get k j expression:

[0026]

[0027] Assume that w j is the weight of the connection between the jth neuron and y, θ y is the bias of y, and there exists a mapping relationship where g is the activation function; the output out expression is:

[0028]

[0029] Finally, out is a four-dimensional vector, corresponding to four different classification states;

[0030] The output layer is mainly composed of a softmax module; assuming that the hidden layer output vector out = (out 1 ,out 2 ,out 3 ,out 4 ), the specific formula is as follows:

[0031]

[0032] Among them, out i represents the output value of the i-th node, and e represents the exponential constant. Finally, the output vector y is obtained:

[0033] y=Softmax(out).

[0034] Preferably, the loss function of the 1DANN fault diagnosis model training is a cross entropy function, as shown below:

[0035]

[0036] Where y is the predicted label of the model, The real label for the molecular pump;

[0037] At the same time, the PSO algorithm is introduced to ensure that the 1DANN neural network avoids local optimal solutions to the greatest extent during the calculation process;

[0038] The PSO algorithm is defined as follows: Assuming that in a D-dimensional search space, a single colony contains N particles, and a single D-dimensional vector is represented by the i-th particle, denoted as:

[0039] X id =(X i1 , X i2 , …, X iD )

[0040] A single D-dimensional vector is the flight speed of the i-th particle, recorded as:

[0041] V id =(V i1 , V i2 , …, V D )

[0042] The individual extreme value is defined as the current optimal position P searched by the i-th particle swarm best , that is, the current local optimal parameter combination scheme, recorded as:

[0043] P best =(P i1 , P i2 ,…,P iD )

[0044] The global extreme value is defined as the optimal position P retrieved by the entire particle swarm d，gbest , that is, the optimal parameter combination scheme in the global case, recorded as:

[0045] P d，gbest =(P 1，gbest , P 2，gbest ,…,P D，gbest )

[0046] When these two optimal values ​​are found, each particle needs to update its speed and position according to the following formula to achieve iterative tuning of parameters. The process is expressed as:

[0047] V id (t+1)=W×V id (t)+C 1 × 1 [P best (t)--X id (t)]+

[0048] C 2 × 2 [P d，gbest (t)-X id (t)]

[0049] X id (t+1)=X id (t)+V id(t+1)

[0050] Where N represents the size of the particle swarm, i (i = 1, 2, ..., N) represents the particle number; D represents the particle dimension; d represents the particle dimension number, d = 1, 2, ..., N; t represents the number of iterations; W represents the inertia weight; C 1 represents the individual learning factor; C 2 represents the group learning factor; r 1 、r 2 Represents the group learning factor, with a value range of [0, 1];

[0051] Among them, the fitness function is the minimum error obtained by five-fold cross validation; the cross validation principle is as follows:

[0052] Divide the data into K groups, extract the training set and validation set in the ratio of K-1 to 1, repeat K times, obtain the status of K groups of training models, and then use relevant evaluation indicators to select the group with the smallest error to define the optimal model.

[0053] A second aspect of the present invention provides a high-precision molecular pump fault diagnosis method, comprising the following process:

[0054] S1, collects the vibration frequency signal of the molecular pump, and the frequency is set to 2.25kHz;

[0055] S2, performing time domain and frequency domain analysis on the molecular pump vibration frequency signal collected in real time to extract the real-time feature vector;

[0056] S3, inputting the real-time feature vector extracted by S2 into the 1DANN fault diagnosis model constructed by the construction method described in the first aspect;

[0057] S4, output the test results and issue a warning if a fault occurs.

[0058] The third aspect of the present invention provides a high-precision molecular pump fault diagnosis device, which includes at least one processor and at least one memory, and the processor and the memory are coupled; the memory stores a computer execution program of the 1DANN fault diagnosis model built by the building method described in the first aspect; when the processor executes the computer execution program stored in the memory, the processor executes a high-precision molecular pump fault diagnosis method.

[0059] The fourth aspect of the present invention provides a computer-readable storage medium, which stores a computer program or instruction of the 1DANN fault diagnosis model constructed by the construction method described in the first aspect. When the program or instruction is executed by a processor, the processor executes a high-precision molecular pump fault diagnosis method.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The present invention uses a neural network optimized by swarm intelligence search for molecular pump fault diagnosis. The method aims to simultaneously ensure high accuracy and effectively solve the problem of model overfitting. First, a new neural network structure, the 1DANN fault diagnosis model, is designed to effectively judge the molecular pump fault diagnosis data. Next, the present invention combines the PSO algorithm on the basis of the 1DANN neural network to ensure that it avoids falling into a local optimal solution during the calculation process. At the same time, combined with the five-fold cross-validation algorithm, model overfitting can be avoided, thereby finding the best fitness value, obtaining the optimal parameters, and training the optimal neural network model, thereby improving the accuracy of molecular pump fault diagnosis.

[0062] The specific implementation of the technical improvements brings unique advantages to each link. The introduction of the 1DANN neural network makes the judgment of molecular pump fault diagnosis data more accurate; the combination of the PSO algorithm can avoid falling into the dilemma of local optimal solutions, ensuring the efficiency and accuracy of the calculation; the application of the five-fold cross-validation algorithm effectively avoids model overfitting and ensures the generalization ability of the model. In summary, these technical improvements work together, so that the present invention can bring a more reliable and efficient solution to molecular pump fault diagnosis, thereby promoting the development and application of related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of the overall scheme of the present invention.

[0064] Figure 2 It is a principle block diagram of the 1DANN neural network fault diagnosis model of the present invention.

[0065] Figure 3 This is the flow chart of the PSO algorithm optimizing the neural network for the swarm intelligent search algorithm of the present invention.

[0066] Figure 4 This is a simplified structural diagram of the high-precision molecular pump fault diagnosis equipment in Example 2. DETAILED DESCRIPTION

[0067] The present invention is further described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] Embodiment 1:

[0069] In the process flow of semiconductor manufacturing, many processes need to be carried out in an ultra-vacuum environment, which requires the use of molecular pump equipment. The flowing gas molecules collide violently with the rotating blades or rotors driven by the motor, causing the gas molecules in the pump to continuously lose kinetic energy and eventually be pumped out of the vacuum chamber, thereby providing a high-vacuum operating environment for the later processes. The bearings at the upper end of the molecular pump rotor are wear-free permanent magnetic bearings, and the lower end is a ceramic ball bearing. The results of reliability experiments show that the most common and most harmful failures of molecular pumps are the failure of ceramic ball bearings and the fatigue of rubber vibration damping pads on ceramic ball bearings. Bearing failure will cause the vibration noise of the molecular pump to increase, the load to increase, and even fail to work properly. Fatigue of the rubber vibration damping pad will cause changes in the axial size and stiffness, affecting the rotor dynamics of the molecular pump.

[0070] The present invention proposes a molecular pump fault diagnosis method and system based on swarm intelligent search and optimization neural network through time domain and frequency domain analysis, which can effectively solve the problem of model overfitting while ensuring high accuracy. First, the present invention proposes a 1DANN neural network to effectively judge the molecular pump fault diagnosis data. Secondly, the present invention introduces the PSO algorithm on the basis of the original 1DANN neural network to ensure that 1DANN can avoid local optimal solutions to a large extent during the calculation process. At the same time, it combines the five-fold cross-validation algorithm to avoid model overfitting, thereby finding the best fitness value, obtaining the optimal parameters, and training the optimal neural network model to improve the accuracy of molecular pump fault diagnosis. The overall process is as follows: Figure 1 shown.

[0071] This embodiment first provides a method for building a high-precision molecular pump fault diagnosis model, including the following process:

[0072] Step 1, collecting vibration signals of the molecular pump under different states, including vibration signals of a ceramic ball bearing failure state, a rubber vibration damping pad on the ceramic ball bearing slightly fatigued state, a rubber vibration damping pad severely fatigued state, and a normal state;

[0073] Step 2, performing time domain and frequency domain analysis on the collected data to obtain a feature-extracted data set based on the time domain and frequency domain analysis;

[0074] Step 3, selecting frequency domain features of different states and combining them with time domain features to construct a vibration feature data set of different states of the molecular pump, and dividing the feature data set into a training set and a test set;

[0075] Step 4, design a 1DANN fault diagnosis model based on swarm intelligence search optimization PSO algorithm and train it using the feature data set;

[0076] Step 5: Use the test set to test the trained 1DANN fault diagnosis model and obtain the final 1DANN fault diagnosis model.

[0077] 1. Data Collection

[0078] In order to collect molecular pump fault signals, this embodiment collects one ceramic ball bearing that failed in the full life experiment, one rubber vibration damping pad with slight fatigue, and one rubber vibration damping pad with severe fatigue. By replacing parts, experiments were carried out under four conditions: bearing failure (category 1), slight fatigue of vibration damping pad (category 2), severe fatigue of vibration damping pad (category 3), and normal working condition (category 4). The sampling frequency is set to 1kHz. When collecting experimental data, the peak-to-peak value of the signal generator is set to 15V, the frequency is set to 2.25kHz, the signal is manually triggered, the number of cycles is set to 9000, and 20,000 normal data are collected through multiple triggering. Replace the ceramic ball bearing and collect 8,000 category 1 fault data. Replace the rubber vibration damping pad with slight fatigue and collect 6,000 category 2 fault data. Replace the rubber vibration damping pad with severe fatigue and collect 4,000 category 2 fault data. Finally, the molecular pump fault diagnosis data set X is obtained, and the sampling period is 0.001s.

[0079] 2. Time domain and frequency domain analysis

[0080] For the collected sample data X, assuming that N represents the number of sampling points collected in a period, x i represents the i-th data sample in the current cycle, then i∈[1,N]. In the time domain analysis, the maximum value X is selected. max =max(x i ), minimum value X min =min(x i ), peak-to-peak value X ff =X max -X min ,average value variance RMS Crest Factor Pulse Factor As a vibration state analysis, to better reflect the hidden characteristics of the data, in the frequency domain analysis, the time domain signal is converted into the frequency domain signal through the fast Fourier transform (FFT), and the FFT formula is:

[0081]

[0082] Where k = 0, 1, ..., N-1 represents the signal length, is the rotation factor, f k is the frequency domain value, x(n) is the sampling point in the time domain, n is the sequence index of the time domain sampling point, and k is the index of the frequency domain value. In the present invention, the frequency domain mean is selected Frequency Domain Center of Gravity Frequency domain standard deviation The three-state frequency domain characteristics describe the vibration state information of the molecular pump.

[0083] The data x i The maximum value, minimum value, peak-to-peak value, average value, variance, root mean square, peak factor, pulse factor in the time domain analysis are combined with the frequency domain mean, frequency domain centroid, and frequency domain standard deviation in the frequency domain analysis to form a feature vector f i , the final input sampling data set X becomes the feature extracted data set F based on time domain and frequency domain analysis.

[0084] For the obtained dataset F after feature extraction based on time domain and frequency domain analysis, every 10 data are sliced ​​into a group of samples, and 2000 groups of normal vibration feature data, 800 groups of ceramic ball bearing failure data, 600 groups of slightly fatigued rubber vibration damping pad failure vibration feature data, and 400 groups of severely fatigued rubber vibration damping pad failure vibration feature data are obtained. The training set and the test set are divided into 8:2 to obtain the molecular pump vibration training set F. train With the test set F val .

[0085] 3.1 Construction of DANN neural network fault diagnosis model

[0086] The calculation and learning within the 1DANN neural network can generally be divided into two steps in different directions: forward propagation of information and reverse adjustment of errors. The first is the forward propagation of information, that is, the process in which information is transmitted through the input layer to the hidden layer and finally output by the output layer. At this time, if the actual output is the same as the expected output, the learning and training ends here; if the result does not match the expected value, the error signal will return to the original path, and the error signal will decrease along the negative gradient direction by modifying the connection weights of the output layer and the hidden layer in turn, that is, adjusting the weights to reduce the error. The 1DANN neural network mainly includes an input layer, an output layer, and a hidden layer, such as Figure 2 shown.

[0087] The input layer of the 1DANN neural network is mainly composed of a fully connected layer, the main purpose of which is to input the molecular pump training data set F train The 11-dimensional features are mapped to the 120-dimensional latent function feature space. The main formula of the input layer is as follows:

[0088]

[0089] in is the feature map matrix.

[0090] The hidden layer of 1DANN neural network is as follows: is the hidden layer input variable, y is the output variable, k m For hidden neuron output, there exists a mapping relationship where g is the activation function. Assume vi,j is the weight of the i-th input variable and the j-th hidden layer neuron: is the threshold of the jth neuron in the hidden layer k, i.e., the bias term. j expression:

[0091]

[0092] Assume that w j is the weight of the connection between the jth neuron and y, θ y is the bias of y, and there exists a mapping relationship where g is the activation function. The output out expression is:

[0093]

[0094] Finally, out is a four-dimensional vector, corresponding to four different classification states.

[0095] The 1DANN neural network mainly includes an output module composed of a softmax module. Assume that the hidden layer output vector out = (out 1 ,out 2 ,out 3 ,out 4 ), the specific formula is as follows:

[0096]

[0097] Among them, out i represents the output value of the i-th node, and e represents the exponential constant. Finally, the output vector y is obtained:

[0098] y=Softmax(out)

[0099] The hyperparameters of the network are defined as shown in Table 1:

[0100] Table 11 DANN network parameter settings

[0101] parameter set up parameter set up Input layer dimensions 120 Target error 0.00001 Hidden layer dimensions 240 Maximum number of training sessions 500 Output layer dimension 4 Maximum number of iterations 75 Learning Rate 0.002 Population size 8

[0102] 4. Design a 1DANN fault diagnosis model based on swarm intelligence search optimization PSO algorithm

[0103] The loss function of 1DANN neural network training is the cross entropy function, as shown below:

[0104]

[0105] Where y is the predicted label of the model, is the true label of the molecular pump. However, in the process of 1DANN neural network training, there is often a problem of falling into the local optimal solution, and it is not easy to train the optimal model.

[0106] In order to solve this problem, the present invention proposes to introduce the PSO algorithm on the basis of the original 1DANN neural network to ensure that the 1DANN neural network can avoid local optimal solutions to a large extent during the calculation process. For the PSO algorithm, a particle in the group, i.e., the parameters of the input layer, output layer, and hidden layer in the 1DANN neural network, each particle corresponds to a set of parameter combination schemes (candidate solutions) for the overall optimization problem, and the optimal parameter combination (optimal solution) in the search space can be explored through iteration.

[0107] The PSO algorithm optimization flow chart is as follows Figure 3 As shown: Assuming that in a D-dimensional search space, a single colony contains N particles, a single D-dimensional vector is represented by the i-th particle, which is recorded as:

[0108] X id =(X i1 , X i2 , …, X iD )

[0109] A single D-dimensional vector is the flight speed of the i-th particle, recorded as:

[0110] V id =(V i1 , V i2 , …, V iD )

[0111] The individual extreme value is defined as the current optimal position P searched by the i-th particle swarm best , that is, the current local optimal parameter combination scheme, recorded as:

[0112] P best =(P i1 , P i2 ,…,P iD )

[0113] The global extreme value is defined as the optimal position P retrieved by the entire particle swarm d，gbest , that is, the optimal parameter combination scheme in the global case, recorded as:

[0114] P d，gbest =(P 1，gbest , P 2，gbest ,…,P D，gbest )

[0115] When these two optimal values ​​are found, each particle needs to update its speed and position according to the following formula to achieve iterative tuning of parameters. The process is expressed as:

[0116] V id (t+1)=W×V id (t)+C1 × 1 [P best (t)-X id (t)]+

[0117] C 2 × 2 [P d，gbest (t)-X id (t)]

[0118] X id (t+1)=X id (t)+V id (t+1)

[0119] Where N represents the size of the particle swarm, i (i = 1, 2, ..., N) represents the particle number, D represents the particle dimension, d (d = 1, 2, ..., N) represents the particle dimension number, t represents the number of iterations, W represents the inertia weight, C represents the particle size, and C represents the particle size. 1 represents the individual learning factor; C 2 represents the group learning factor; r 1 、r 2 Represents the group learning factor, and its value range is [0,1]; the fitness function is the minimum error obtained by five-fold cross validation. The principle of cross validation is as follows: divide the data into K groups, extract the training set and the validation set in the ratio of K-1 to 1, repeat K times, obtain the K groups of training model states, and then use relevant evaluation indicators to select the group with the smallest error to define the optimal model. The present invention selects five-fold cross validation to avoid overfitting in model training to improve the prediction accuracy of the model. The particle swarm algorithm has the advantages of faster convergence speed and efficient global search capability, and is insensitive to the initialization parameters of the 1DANN neural network. The process is as follows Figure 2 As shown in the figure, PSO helps to improve the performance and generalization ability of 1DANN neural network, thereby better improving the accuracy of molecular pump fault diagnosis.

[0120] Embodiment 2:

[0121] like Figure 4As shown, the present invention also provides a high-precision molecular pump fault diagnosis device, the device includes at least one processor and at least one memory, and also includes a communication interface and an internal bus; the memory stores a computer execution program of the 1DANN fault diagnosis model built by the construction method described in Example 1; when the processor executes the computer execution program stored in the memory, the processor can execute a high-precision molecular pump fault diagnosis method. The internal bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus. The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a U disk, a mobile hard disk, a read-only memory, a disk or an optical disk.

[0122] The device may be provided as a terminal, a server or other forms of devices.

[0123] Figure 4 is a block diagram of a device shown as an example. The device may include one or more of the following components: a processing component, a memory, a power component, a multimedia component, an audio component, an input / output (I / O) interface, a sensor component, and a communication component. The processing component generally controls the overall operation of the electronic device, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component may include one or more processors to execute instructions to complete all or part of the steps of the above method. In addition, the processing component may include one or more modules to facilitate interaction between the processing component and other components. For example, the processing component may include a multimedia module to facilitate interaction between the multimedia component and the processing component.

[0124] The memory is configured to store various types of data to support operations on the electronic device. Examples of such data include instructions for any application or method operating on the electronic device, contact data, phone book data, messages, pictures, videos, etc. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0125] The power supply assembly provides power to various components of the electronic device. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device. The multimedia assembly includes a screen that provides an output interface between the electronic device and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia assembly includes a front camera and / or a rear camera. When the electronic device is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0126] The audio component is configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), and when the electronic device is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in a memory or sent via a communication component. In some embodiments, the audio component also includes a speaker for outputting an audio signal. The I / O interface provides an interface between the processing component and the peripheral interface module, and the above-mentioned peripheral interface module can be a keyboard, a click wheel, a button, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

[0127] The sensor assembly includes one or more sensors for providing various aspects of status assessment for the electronic device. For example, the sensor assembly can detect the on / off state of the electronic device, the relative positioning of components, such as the display and keypad of the electronic device, and the sensor assembly can also detect the position change of the electronic device or a component of the electronic device, the presence or absence of user contact with the electronic device, the orientation or acceleration / deceleration of the electronic device and the temperature change of the electronic device. The sensor assembly may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.

[0128] The communication component is configured to facilitate wired or wireless communication between the electronic device and other devices. The electronic device can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0129] In an exemplary embodiment, the electronic device may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above methods.

[0130] Embodiment 3:

[0131] The present invention also provides a computer-readable storage medium, which stores a computer program or instruction of the 1DANN fault diagnosis model constructed by the construction method described in Example 1. When the program or instruction is executed by a processor, the processor can execute a high-precision molecular pump fault diagnosis method.

[0132] Specifically, a system, device or equipment equipped with a readable storage medium may be provided, on which a software program code for implementing the functions of any of the above-mentioned embodiments is stored, and a computer or processor of the system, device or equipment reads and executes the instructions stored in the readable storage medium. In this case, the program code read from the readable medium itself can implement the functions of any of the above-mentioned embodiments, so the machine-readable code and the readable storage medium storing the machine-readable code constitute a part of the present invention.

[0133] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk (such as CD-ROM, CD-R, CD-RW, DVD-20ROM, DVD-RAM, DVD-RW, DVD-RW), magnetic tape, etc. The storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0134] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0135] It should be understood that the storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in a terminal or server.

[0136] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0137] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0138] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0139] Although the above describes the specific implementation methods of the present invention, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for building a high-precision molecular pump fault diagnosis model, characterized in that: The process includes: Step 1, collecting vibration signals of the molecular pump under different states, including the failure state of the ceramic ball bearing, the slight fatigue state of the rubber vibration damping pad on the ceramic ball bearing, the severe fatigue state of the rubber vibration damping pad and the vibration signal of the normal state; specifically: The ceramic ball bearings that failed in the full life experiment, the rubber vibration damping pads of the ceramic ball bearings with slight fatigue and the rubber vibration damping pads of the ceramic ball bearings with severe fatigue were collected. The experiments were carried out under four conditions: ceramic ball bearing failure category 1, rubber vibration damping pad slight fatigue category 2, rubber vibration damping pad severe fatigue category 3 and normal working condition category 4 by replacing parts. The sampling frequency was set to 1kHz. When collecting experimental data, the peak-to-peak value of the signal generator was set to 15V, the frequency was set to 2.25kHz, the signal was manually triggered, the number of cycles was set to 9000, and 20,000 normal data were collected through multiple triggering. The ceramic ball bearings were replaced to collect 8,000 category 1 fault data; the rubber vibration damping pads with slight fatigue were replaced to collect 6,000 category 2 fault data; the rubber vibration damping pads with severe fatigue were replaced to collect 4,000 category 2 fault data, and finally the molecular pump fault diagnosis data set was obtained. , the sampling period is 0.001s; Step 2, performing time domain and frequency domain analysis on the collected data to obtain a feature-extracted data set based on the time domain and frequency domain analysis; Specifically: For the collected sample data , assuming Represents the number of sampling points collected in a cycle. Indicates the current cycle data samples, then ; Selecting the maximum value in time domain analysis , minimum , peak-to-peak ,average value ,variance , RMS , Peak Factor , Pulse Factor As a vibration state analysis, it better reflects the hidden characteristics of data; In frequency domain analysis, the time domain signal is converted into a frequency domain signal through the fast Fourier transform FFT. The FFT formula is: in Indicates the signal length, is the rotation factor, is the frequency domain value, is the sampling point in the time domain, is the sequence index of the time domain sampling points, is the index of the frequency domain value; Use frequency domain mean , frequency domain center of gravity , frequency domain standard deviation The three-state frequency domain characteristics describe the vibration state information of the molecular pump; The data The maximum value, minimum value, peak-to-peak value, average value, variance, root mean square, peak factor, pulse factor in the time domain analysis and the frequency domain mean, frequency domain centroid, and frequency domain standard deviation in the frequency domain analysis are combined to form a feature vector , the final input sampling data set It becomes a feature-extracted data set based on time domain and frequency domain analysis ; Step 3, selecting frequency domain features of different states and combining them with time domain features to construct a vibration feature data set of different states of the molecular pump, and dividing the feature data set into a training set and a test set; Step 4, designing a 1DANN fault diagnosis model based on swarm intelligence search optimization PSO algorithm, and training it using a feature data set; the 1DANN fault diagnosis model includes an input layer, an output layer and a hidden layer; The input layer is mainly composed of a fully connected layer, which is used to input the molecular pump training data set The 11-dimensional features are mapped to the 120-dimensional implicit function feature space; the main formula of the input layer is as follows: in is the feature mapping matrix; The hidden layer is specifically, assuming Input variables for the hidden layer, is the output variable, is the hidden neuron output, there exists is the mapping relationship of the activation function; assuming For the The input variable and The weights of the neurons in the hidden layer, For hidden layer No. The threshold of the neuron, that is, the bias term; we can get expression: Assumptions For the Neurons and The weight of the connection, for The bias exists is the mapping relationship of the activation function; the output The expression is: final is a four-dimensional vector, corresponding to four different classification states; The output layer is mainly composed of softmax modules; assuming that the hidden layer output vector , the specific formula is as follows: in, Indicates The output value of each node, represents the exponential constant, and finally the output vector is obtained : Step 5: Use the test set to test the trained 1DANN fault diagnosis model and obtain the final 1DANN fault diagnosis model.

2. A method for building a high-precision molecular pump fault diagnosis model as claimed in claim 1, characterized in that: The step 3 is specifically as follows: Dataset after feature extraction based on time domain and frequency domain analysis , every 10 data are sliced ​​into a group of samples, and 2000 groups of normal vibration feature data, 800 groups of ceramic ball bearing failure data, 600 groups of slightly fatigued rubber vibration damping pad failure vibration feature data, 400 groups of severely fatigued rubber vibration damping pad failure vibration feature data are obtained. The training set and the test set are divided into 8:2 to obtain the molecular pump vibration training set With the test set .

3. A method for building a high-precision molecular pump fault diagnosis model according to claim 1, characterized in that: The loss function of the 1DANN fault diagnosis model training is the cross entropy function, as shown below: in is the predicted label of the model, The real label for the molecular pump; At the same time, the PSO algorithm is introduced to ensure that the 1DANN neural network avoids local optimal solutions to the greatest extent during the calculation process; The PSO algorithm is defined as follows: Assume In the dimensional search space, a single community contains Particles, single The dimension vector is Particles are represented as: single The dimension vector is The flying speed of a particle is recorded as: The individual extreme value is defined as The current optimal position searched by the particle swarm , that is, the current local optimal parameter combination scheme, recorded as: The global extreme value is defined as the optimal position retrieved by the entire particle swarm , that is, the optimal parameter combination scheme in the global case, recorded as: When these two optimal values ​​are found, each particle needs to update its speed and position according to the following formula to achieve iterative tuning of parameters. The process is expressed as: in represents the particle swarm size, represents the particle number; represents the particle dimension; Represents the particle dimension number, ; Represents the number of iterations; represents the inertia weight; represents the individual learning factor; represents the group learning factor; , Represents the group learning factor, with a value range of ; Among them, the fitness function is the minimum error obtained by five-fold cross validation; the cross validation principle is as follows: Divide the data into Group, press and Extract training set and validation set in the ratio of times, obtain The training model status of each group is calculated, and then the group with the smallest error is selected using relevant evaluation indicators to define the optimal model.

4. A high-precision molecular pump fault diagnosis method, characterized in that: The process includes: S1, collect the vibration frequency signal of the molecular pump, the frequency is set to 2.25kHz; S2, performing time domain and frequency domain analysis on the molecular pump vibration frequency signal collected in real time to extract the real-time feature vector; S3, inputting the real-time feature vector extracted by S2 into the 1DANN fault diagnosis model constructed by the construction method according to any one of claims 1 to 3; S4, output the test results and issue a warning if a fault occurs.

5. A high-precision molecular pump fault diagnosis device, characterized in that: The device includes at least one processor and at least one memory, the processor and the memory are coupled; the memory stores a computer execution program of a molecular pump fault diagnosis method; when the processor executes the computer execution program stored in the memory, a high-precision molecular pump fault diagnosis method as described in claim 4 is implemented.

6. A computer-readable storage medium, characterized in that: The computer readable storage medium stores a computer program or instruction of a molecular pump fault diagnosis method. When the program or instruction is executed by a processor, a high-precision molecular pump fault diagnosis method as claimed in claim 4 is implemented.

Images