Vacuum pump fault prediction method and device

Abstract

The invention discloses a vacuum pump fault prediction method and device, and belongs to the technical field of vacuum pumps, and the vacuum pump fault prediction method comprises the steps: collecting the operation data of a vacuum pump; the operation data are input into a mechanism model in the hybrid model, the first vacuum degree of the vacuum pump is obtained, and the mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump; and inputting the operation data and the first vacuum degree into a machine learning model in the hybrid model to obtain a fault prediction result of the vacuum pump, thereby improving the accuracy of fault prediction of the vacuum pump.

Description

Technical Field

[0001] This application relates to the field of vacuum pump technology, specifically to a method and apparatus for predicting vacuum pump failures. Background Technology

[0002] Vacuum pumps are core equipment in manufacturing industries such as semiconductors, chemicals, and pharmaceuticals, used to maintain the vacuum environment for critical processes. Monitoring the condition and predicting the failures of vacuum pumps, and implementing predictive maintenance, are crucial for ensuring production safety and continuous production. However, existing vacuum pump failure prediction solutions suffer from low accuracy. Summary of the Invention

[0003] To address the aforementioned technical problems, embodiments of this application provide a vacuum pump fault prediction method, apparatus, electronic device, computer-readable storage medium, and computer program product, which can improve the accuracy of vacuum pump fault prediction.

[0004] Firstly, a method for predicting vacuum pump failures is provided, comprising the following steps: Collect operating data of the vacuum pump; The operational data is input into the mechanistic model in the hybrid model to obtain the first vacuum degree of the vacuum pump. The mechanistic model is constructed based on the working principle and physical characteristics of the vacuum pump. The operating data and the first vacuum level are input into the machine learning model in the hybrid model to obtain the failure prediction result of the vacuum pump.

[0005] Secondly, a vacuum pump failure prediction device is also provided, the device comprising: The data acquisition module is used to collect the operating data of the vacuum pump; The first prediction module is used to input the operating data into the mechanism model in the hybrid model to obtain the first vacuum degree of the vacuum pump. The mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump. The second prediction module is used to input the operating data and the first vacuum degree into the machine learning model in the hybrid model to obtain the failure prediction result of the vacuum pump.

[0006] Thirdly, embodiments of this application also provide an electronic device, including a processor and a memory, wherein the memory stores multiple instructions; the processor loads instructions from the memory to execute the steps of any of the vacuum pump fault prediction methods provided in embodiments of this application.

[0007] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute the steps of any of the vacuum pump fault prediction methods provided in embodiments of this application.

[0008] Fifthly, embodiments of this application also provide a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps in any of the vacuum pump fault prediction methods provided in embodiments of this application.

[0009] Beneficial effects: In this embodiment, by collecting actual operating data of the vacuum pump, a real and direct data foundation is provided for fault prediction; the mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump, which can theoretically reflect the relationship between operating data and vacuum degree; and the machine learning model comprehensively utilizes the operating data and the first vacuum degree output by the mechanism model to perform fault analysis on the vacuum pump, thereby giving full play to the advantages of the mechanism model in reflecting physical characteristics and utilizing the ability of the machine learning model to identify faults from data, thus improving the accuracy and reliability of fault prediction results. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram illustrating the application environment of the vacuum pump fault prediction method provided in some embodiments of this application; Figure 2 This is a flowchart illustrating a vacuum pump failure prediction method provided in some embodiments of this application; Figure 3 This is a flowchart illustrating the process of determining the target parameter values ​​of the preset operating parameters of the vacuum pump provided in some embodiments of this application; Figure 4 This is another flowchart illustrating the vacuum pump failure prediction method provided in some embodiments of this application; Figure 5 This is another flowchart illustrating a vacuum pump failure prediction method provided in some embodiments of this application; Figure 6 This is a schematic diagram of the structure of a vacuum pump fault prediction device provided in some embodiments of this application; Figure 7 This is a schematic diagram of the internal structure of an electronic device provided in some embodiments of this application. Detailed Implementation

[0012] The following describes the relevant content, terms, meanings, technical issues, technical solutions, and beneficial effects involved in the embodiments of this application.

[0013] In industrial production, vacuum pumps are core equipment for maintaining a vacuum environment in many industries such as electronics, chemicals, and food. As operating time accumulates, even if a vacuum pump appears to be functioning normally, its internal components will gradually experience irreversible wear and aging, eventually leading to failure.

[0014] Currently, existing technologies for vacuum pump operation control and fault prediction have the following shortcomings: First, in setting operating parameters, they often rely on empirical values ​​or fixed ranges, failing to fully consider the impact of dynamic changes in the cooling system flow rate on the vacuum pump's temperature and overall energy consumption, and lacking a dynamic optimization mechanism based on real-time status. Second, in fault diagnosis, traditional methods often rely on a single threshold (such as temperature) or simple statistical analysis, making it difficult to accurately identify specific fault types; in some cases, even manual inspection is still used, relying on subjective means such as observing the vacuum pump's sound and vibration for judgment, which limits both efficiency and accuracy. Furthermore, vibration and temperature signals collected in industrial settings are easily affected by electromagnetic interference and mechanical noise, and existing deep learning models are quite sensitive to data disturbances, often leading to significant fluctuations in prediction results.

[0015] To address the aforementioned issues, this application provides a vacuum pump fault prediction method, a vacuum pump fault prediction device, an electronic device, a computer-readable storage medium, and a computer program product. In this application, by collecting actual operating data of the vacuum pump, a real and direct data foundation is provided for fault prediction. The mechanistic model, constructed based on the working principle and physical characteristics of the vacuum pump, can theoretically reflect the correlation between operating data and vacuum level. The machine learning model comprehensively utilizes the operating data and the first vacuum level output by the mechanistic model to perform fault analysis on the vacuum pump. This leverages both the advantages of the mechanistic model in reflecting physical characteristics and the machine learning model's ability to identify faults from data, thereby improving the accuracy and reliability of the fault prediction results.

[0016] To better understand the vacuum pump fault prediction method, vacuum pump fault prediction device, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of this application, the application environment applicable to the embodiments of this application is described below.

[0017] Please see Figure 1 , Figure 1 This diagram illustrates an application environment for a vacuum pump fault prediction method provided in an embodiment of this application. As one implementation, the vacuum pump fault prediction method provided in this embodiment can be applied to an electronic device. This electronic device can be, for example,... Figure 1The server 110 shown can be connected to the terminal device 120 via a network. The network serves as a medium for providing a communication link between the server 110 and the terminal device 120. The network can include various connection types, such as wired communication links, wireless communication links, etc., and this embodiment is not limited thereto. Optionally, in other embodiments, the electronic device can also be a smartphone, laptop, etc.

[0018] It should be understood that Figure 1 The server 110, network, and terminal device 120 shown are merely illustrative. Depending on the implementation requirements, any number of servers, networks, and terminal devices can be included. For example, server 110 can be a physical server or a server cluster consisting of multiple servers, and terminal device 120 can be a mobile phone, tablet, desktop computer, laptop computer, smart speaker, smart wearable device, etc. It is understood that embodiments of this application can also allow multiple terminal devices 120 to access server 110 simultaneously.

[0019] In some embodiments, the terminal device 120 may send a vacuum pump fault prediction request to the server. After receiving the vacuum pump fault prediction, the server 110 may perform fault prediction using the vacuum pump fault prediction method described in the embodiments of this application.

[0020] As another implementation, the server 110 and the terminal device 120 described in this application embodiment can be integrated, such as the server 110 or the terminal device 120 directly receiving the user's input vacuum pump fault prediction request and performing fault prediction.

[0021] The following describes in detail, with reference to the accompanying drawings, the vacuum pump fault prediction method, vacuum pump fault prediction device, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of this application. It should be noted that the order of description of the following embodiments is not intended to limit the preferred order of embodiments. Although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be performed in a different order than that shown in the accompanying drawings.

[0022] On the one hand, this embodiment provides a method for predicting vacuum pump failures, such as... Figure 2 As shown, it includes the following steps S21-S23: S21. Collect the operating data of the vacuum pump.

[0023] Specifically, sensors deployed on the vacuum pump and its associated systems (e.g., cooling systems) to be fault-predicted can collect multi-source operating data reflecting the operating status of the vacuum pump in real time or at a set sampling frequency (e.g., once per second). This operating data may include at least a portion of the following: current and voltage of the vacuum pump motor collected by current and voltage sensors; real-time speed of the vacuum pump motor collected by speed sensors; power values ​​obtained by power meters or calculated based on electrical parameters; temperature data collected by temperature sensors installed on the motor housing, bearing housing, pump body surface, and cooling water inlet and outlet; flow rate of the cooling system collected by flow meters installed in the cooling pipeline; inlet and outlet pressure (or vacuum degree) data collected by pressure sensors or vacuum gauges installed at the vacuum pump inlet and outlet; and vibration data collected by vibration sensors (e.g., accelerometers) installed on the pump body or base.

[0024] Among these, current and voltage are related to the motor load of the vacuum pump and can be used to determine whether the motor is operating normally; temperature data is related to motor overload or cooling system failure; and inlet and outlet pressures are related to the pumping performance and sealing condition of the vacuum pump. The actual operating data of the vacuum pump provides a data foundation for subsequent fault prediction.

[0025] S22. Input the operating data into the mechanism model in the hybrid model to obtain the first vacuum degree of the vacuum pump. The mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump.

[0026] In this embodiment, the hybrid model includes a mechanistic model and a machine learning model. The mechanistic model is constructed based on the working principle and physical characteristics of the vacuum pump. For example, the mechanistic model is a mathematical model established based on the physical principles of gas dynamics, thermodynamics and fluid dynamics of the vacuum pump. The mechanistic model describes the performance state that the vacuum pump should theoretically achieve under given operating conditions (e.g., speed, cooling conditions, intake load). This performance state may include the vacuum state.

[0027] For example, a mechanism model can be obtained by establishing a mathematical relationship between the motor speed N of the vacuum pump, the flow rate Q of the cooling medium (e.g., cooling water or cooling gas), and the vacuum degree Ppred. In one specific embodiment, this mechanism model can be expressed as: Ppred=f(N,Q) Where Ppred is the first vacuum level calculated by the mechanistic model, N is the motor speed of the vacuum pump, Q is the cooling medium flow rate of the vacuum pump, and f(N,Q) is the function expression corresponding to the mechanistic model. The inputs to this function expression include the motor speed N and the cooling medium Q. The function expression of f(N,Q) can be determined according to the type and structural parameters of the vacuum pump. Its construction process involves equations such as mass conservation and energy balance. Therefore, there can be one mechanistic model for each type of vacuum pump.

[0028] In this embodiment, the type of vacuum pump to be predicted is determined, the corresponding mechanism model is obtained, and the motor speed N and cooling medium flow rate Q from the collected operating data are input into the mechanism model to output the first vacuum degree Ppred of the vacuum pump. The first vacuum degree Ppred reflects the vacuum level that the vacuum pump should reach under theoretical conditions and current operating conditions.

[0029] S23. Input the operating data and the first vacuum degree into the machine learning model in the hybrid model to obtain the failure prediction result of the vacuum pump.

[0030] In this step, the machine learning model (e.g., BP neural network, support vector machine, random forest, etc.) is a pre-trained intelligent model capable of fault identification and classification. Its input data includes not only the operational data collected in step S21, but also the first vacuum degree calculated by the mechanism model in step S22. The machine learning model comprehensively analyzes the input data through the nonlinear mapping relationship learned internally, and the model's output is the fault prediction result.

[0031] In this embodiment, the first vacuum level output by the mechanistic model based on the physical principles of the vacuum pump is a physically meaningful feature. When this feature is input together with the multi-source operating data of the vacuum pump into the machine learning model, it provides a reference benchmark representing the theoretically expected state. By learning the correlation between operating data (e.g., the actual vacuum level in the operating data) and the first vacuum level, and the correlation between operating data and fault types, the machine learning model can gain a deeper understanding of the operating rules of the vacuum pump. Therefore, by combining a hybrid model that integrates physical principles and machine learning capabilities, fault symptoms can be captured more comprehensively, earlier, and more accurately, improving the accuracy and reliability of fault prediction.

[0032] In some embodiments, the fault prediction result includes a state type, which includes a normal state, a potential fault state, and a fault state; when the state type is a potential fault state or a fault state, the fault prediction result also includes a corresponding fault type, which includes at least one of mechanical fault, electrical fault, and performance fault.

[0033] Specifically, a normal state refers to the vacuum pump operating healthily without any signs of malfunction. A potential malfunction state refers to the vacuum pump beginning to deviate from its normal state or exhibiting minor abnormalities, but not yet reaching the point of functional failure. A malfunction state refers to the vacuum pump exhibiting a clear abnormality, with one or more functions failing or experiencing performance degradation.

[0034] When the state type is a potential fault state or a fault state, the fault prediction result also includes the fault type, which includes at least one of mechanical faults (e.g., bearing or rotor wear), electrical faults (e.g., component failure), and performance faults (e.g., pipe leaks or cooling failure). In some embodiments, when the state type is a potential fault state or a fault state, the fault prediction result also includes the fault location. For example, one fault prediction result is: the state type is a potential fault state, the fault type is an electrical fault, and the fault location is C11 of circuit board 1.

[0035] In some embodiments, the method further includes: issuing a warning signal based on a preset warning rule when the state type of the fault prediction result is a potential fault state or a fault state.

[0036] Specifically, the preset warning rules define the warning methods (e.g., email warning, SMS warning, or indicator light warning) and warning scope (e.g., warning to the corresponding maintenance personnel, warning to the corresponding maintenance personnel and their supervisor, or warning to the entire maintenance team) for each type of potential fault state or fault state.

[0037] Upon receiving the fault prediction result, if the status type is a potential fault status or a fault status, the target warning method and target warning range are determined based on the target fault type in the fault prediction result and the preset warning rules, and a warning signal is issued based on the target warning method and target warning range.

[0038] At the same time, the operating data and fault prediction results of the vacuum pump can be stored to provide data support for subsequent fault analysis and model optimization.

[0039] In this embodiment, based on the fault prediction results, the potential faults of the vacuum pump are accurately captured, avoiding missed and false alarms. The faults can be detected in time at the incipient stage, reducing production interruptions and economic losses caused by vacuum pump failures.

[0040] In some embodiments, the training process of the machine learning model includes steps S231-S233: S231. Divide the historical operating data of the sample vacuum pump into a training set, a validation set, and a test set, wherein the historical operating data has fault labels; the proportion of the same fault label is the same in the historical operating data, the training set, the validation set, and the test set.

[0041] Specifically, the sample vacuum pumps can be multiple (e.g., 3) vacuum pumps of the same model as the vacuum pump to be predicted for failure. The operating time of each vacuum pump can be different; for example, the operating time of sample vacuum pump 1 can be 2 years, sample vacuum pump 2 can be 3 years, and sample vacuum pump 3 can be 4 years. Historical operating data of these sample vacuum pumps is acquired, covering their entire life cycle. The historical operating data can be divided into multiple data points according to a preset time interval (e.g., 10 minutes). Each data point includes the corresponding vacuum pump's current, voltage, speed, power value, temperature, flow rate, inlet pressure, outlet pressure, vibration data, etc. Each data point is used as a sample, and the average value of each parameter in each data point is used as the corresponding sample value to eliminate high-frequency noise.

[0042] Each sample has a fault label, which includes corresponding fault association information, including status type, fault type, and fault location.

[0043] After obtaining the samples, they are preprocessed, including data cleaning and normalization. Data cleaning removes noise and outliers from the data, which can be done by using statistical methods to identify and remove data points that deviate from the normal range. Normalization maps data of different ranges and magnitudes to a unified interval to improve the training efficiency and stability of the model.

[0044] In this embodiment, the StratifiedShuffleSplit method is used to perform stratified sampling on the preprocessed samples corresponding to the historical operating data. This ensures that the proportion of the same fault label in the resulting training set, validation set, and test set is the same as that in the historical operating data. For example, if the proportion of normal state, potential fault state, and fault state samples in the preprocessed samples corresponding to the historical operating data is 6:2.5:1.5, then the proportion of normal state, potential fault state, and fault state samples in the training set, validation set, and test set will all be 6:2.5:1.5. After stratified sampling, a chi-square test can be used to verify that there is no significant difference between the data distribution in the training set, validation set, and test set and the data distribution in the historical operating data.

[0045] S232. The machine learning model is iteratively trained using the training set, and the performance of the machine learning model is verified using the validation set during the iterative training process. The hyperparameters of the machine learning model are adjusted based on the verification results to obtain an optimized machine learning model.

[0046] In this embodiment, before iteratively training the machine learning model using the training set, a machine learning model is constructed, and the number of nodes in the input layer, hidden layer, and output layer of the machine learning model is determined. The input layer nodes are used to input the parameter values ​​of each parameter in the sample, such as current, voltage, rotational speed, power value, temperature, flow rate, intake pressure, exhaust pressure, and vibration data. The output layer nodes are used to output the fault prediction results, such as state type, fault type, and fault part. The hidden layer nodes are used to perform nonlinear transformation and feature abstraction on the data input by the input layer nodes. The number of each node can be determined through empirical formulas and multiple experimental debugging.

[0047] Then, the machine learning model is iteratively trained using the training set. The weights and thresholds of the machine learning model are adjusted through the backpropagation algorithm so that the output of the machine learning model is as close as possible to the fault label. The mean squared error can be used as the loss function to measure the error between the prediction result and the label.

[0048] During iterative training, the performance of the machine learning model is validated using a validation set (e.g., the model accuracy is calculated based on the number of correctly predicted samples in the validation set and the total number of samples, and the loss value is calculated based on the samples in the validation set). The hyperparameters of the machine learning model are then adjusted based on the validation results to obtain an optimized machine learning model.

[0049] S233. The optimized machine learning model is evaluated using the test set. If the performance evaluation is passed, the optimized machine learning model is determined as the machine learning model that has been trained.

[0050] Specifically, performance evaluation metrics may include overall accuracy, precision and recall for various types of faults, etc. If the performance evaluation results meet the preset pass criteria (e.g., the overall accuracy is higher than the accuracy threshold and the recall of fault types is not lower than the recall threshold), the optimized machine learning model is determined as a trained machine learning model that can be used for actual fault prediction.

[0051] In some embodiments, adjusting the hyperparameters of the machine learning model based on the validation results includes steps S2321-S2324: S2321. During the iterative training process, obtain the first performance index parameter of the machine learning model trained based on the training set; The primary performance metric could be training loss and / or training accuracy.

[0052] S2322. Obtain the second performance index parameters of the machine learning model adjusted based on the validation set; The second performance metric could be validation loss and / or validation accuracy.

[0053] S2323. When the difference between the first performance index parameter and the second performance index parameter exceeds a preset threshold, adjust the hyperparameters of the machine learning model. In this embodiment, adjusting the hyperparameters of the machine learning model includes increasing the regularization constraint coefficient used to calculate the loss function of the machine learning model, wherein the regularization constraint coefficient is used to suppress the excessive growth of the model parameters of the machine learning model.

[0054] S2324. Continue training the machine learning model based on the adjusted hyperparameters.

[0055] In some embodiments, prior to acquiring the operating data of the vacuum pump, the method further includes steps S201-S202: S201. By optimizing the fitness function, the target parameter values ​​of the preset operating parameters of the vacuum pump are determined. The fitness function represents the mapping relationship between the parameter values ​​corresponding to the preset operating parameters and the fitness value corresponding to the vacuum pump.

[0056] Specifically, the fitness function establishes a mapping relationship between the parameter values ​​of preset operating parameters and the comprehensive evaluation index of the vacuum pump (i.e., the fitness value). The number of preset operating parameters can be one or more. For example, if the predicted operating parameters are the motor speed of the vacuum pump and the cooling medium flow rate of the cooling system, then the fitness function indicates the mapping relationship between the motor speed of the vacuum pump and the cooling medium flow rate of the cooling system and the fitness value.

[0057] In this embodiment, the fitness function can be optimized based on the comprehensive evaluation index to be optimized. For example, if the comprehensive evaluation index to be optimized is the error between the first vacuum degree and the target vacuum degree, then the optimized fitness function is to minimize the fitness function. The process of minimizing the fitness function is to find the target parameter values ​​corresponding to the motor speed and the cooling medium flow rate of the cooling system when the fitness value is minimized.

[0058] If the comprehensive evaluation index is the volume of gas extracted per unit time, then the optimal fitness function is to maximize the fitness function. Maximizing the fitness function involves finding the target parameter values ​​corresponding to the motor speed and the cooling medium flow rate of the cooling system when the fitness value is maximized.

[0059] The optimization of the fitness function can be implemented using various optimization algorithms. The optimization algorithm uses preset operating parameters as variables to be optimized and the fitness function as the evaluation criterion. It updates and iterates the corresponding parameter values ​​within the allowed range of preset operating parameters, and finally converges to a set of parameter values ​​that minimizes or maximizes the fitness value. This set of parameter values ​​is then determined as the target parameter values ​​of the preset operating parameters.

[0060] S202. Adjust the vacuum pump based on the target parameter value.

[0061] Specifically, the target parameter value of the preset operating parameters is input to the controller of the vacuum pump. The controller adjusts the corresponding actuator to stabilize the actual parameter value of the preset operating parameters of the vacuum pump near the target parameter value. For example, the motor speed is adjusted by adjusting the motor driver of the vacuum pump to make the motor speed reach the corresponding target parameter value; the flow rate of the cooling medium in the cooling system is adjusted by adjusting the valve of the cooling system of the vacuum pump to reach the corresponding target parameter value.

[0062] Subsequently, the vacuum pump starts operating based on the adjusted target parameter values. This step ensures that the vacuum pump operates in its optimal state from the start-up stage, which can improve the operating efficiency of the vacuum pump, extend the life of components, and also provide a guarantee for collecting operating data of the vacuum pump that truly reflects the optimized operating conditions.

[0063] In some embodiments, determining the target parameter values ​​of the preset operating parameters of the vacuum pump by optimizing the fitness function includes steps S2011-S2014: S2011. Based on the position of each particle in the particle swarm, the second vacuum degree of each particle is determined using the aforementioned mechanism model. The particle corresponds to the preset operating parameters of the vacuum pump, and the position of the particle indicates the parameter value of the preset operating parameters. In this embodiment, the parameter values ​​of the preset operating parameters can be optimized and adjusted by using an improved particle swarm optimization algorithm with the error between the predicted second vacuum degree and the target vacuum degree of the vacuum pump as the optimization target.

[0064] First, the particle swarm is initialized, including optimizing the number of particles in the swarm, the maximum number of particle update iterations, and initializing the position and velocity of the particles. Each particle corresponds to a preset operating parameter (if there are multiple preset operating parameters, each particle corresponds to a parameter group composed of these preset operating parameters), and the particle's position corresponds to the parameter value of the preset operating parameter.

[0065] If there are 100 particles in the particle swarm, and the preset operating parameters are motor speed and cooling system cooling medium flow rate, then after initialization, a vector is obtained by constructing 100 sets of parameter values ​​for motor speed and cooling system cooling medium flow rate.

[0066] Then, the parameters of the motor speed and the cooling medium flow rate of the cooling system are input into the mechanism model to obtain the predicted second vacuum degree.

[0067] S2012. Based on the second vacuum degree, the target vacuum degree, and the fitness function, determine the fitness value of each particle. The fitness value output by the fitness function is positively correlated with the difference between the second vacuum degree and the target vacuum degree. The target vacuum degree is the vacuum degree of the vacuum pump under ideal vacuum conditions. Since the optimization objective is the error between the second vacuum level and the target vacuum level of the vacuum pump, the fitness value output by the fitness function is positively correlated with the difference between the second vacuum level and the target vacuum level. That is, the fitness function can be a function that calculates the difference between the second vacuum level and the target vacuum level based on the parameter values ​​of the preset operating parameters.

[0068] In some embodiments, the fitness function can be calculated using the following formula: F=f(N,Q)-H Where F is the fitness value, N is the motor speed of the vacuum pump, Q is the cooling medium flow rate of the vacuum pump, f(N,Q) is the function expression corresponding to the mechanism model, and H is the target vacuum level.

[0069] S2013. Update the position of each particle based on the fitness value; The fitness value reflects the difference between the second vacuum level and the target vacuum level. Therefore, by minimizing the difference between the second vacuum level and the target vacuum level, the parameter values ​​of the preset operating parameters can be optimized (optimizing the position of each particle in the particle swarm).

[0070] S2014. Based on the updated positions of each particle, determine the target parameter value of the preset operating parameters.

[0071] After the maximum number of update iterations is reached, the position of the particle whose fitness value is at its minimum is determined, and the parameter value corresponding to the position of that particle is used as the target parameter value of the preset running parameters.

[0072] In some embodiments, the fitness value of the fitness function is also positively correlated with the energy loss value of the vacuum pump. Before determining the fitness value of each particle based on the second vacuum degree, the target vacuum degree of the vacuum pump, and the fitness function, the method further includes: for each particle, determining the energy loss value of the vacuum pump at the parameter value indicated by the position of the particle. The step of determining the fitness value of each particle based on the second vacuum degree, the target vacuum degree of the vacuum pump, and the fitness function includes: determining the fitness value of each particle based on the energy loss value, the second vacuum degree, the target vacuum degree, and the fitness function.

[0073] Specifically, the optimization objective can be the difference between the second vacuum level and the target vacuum level, as well as the energy loss value of the vacuum pump. Therefore, the fitness value output by the fitness function is positively correlated with the difference between the second vacuum level and the target vacuum level, and also positively correlated with the energy loss value of the vacuum pump. That is, the formula for calculating the fitness function can include: the difference between the second vacuum level and the target vacuum level calculated based on preset operating parameter values, and the calculated energy loss value (which can be used as a penalty).

[0074] In some embodiments, the expression for the fitness function can be: F=f(N,Q)-H+E(N,Q) Where F is the fitness value, N is the motor speed of the vacuum pump, Q is the cooling medium flow rate of the vacuum pump, f(N,Q) is the function expression corresponding to the mechanism model, E(N,Q) is the function expression corresponding to the energy loss, and H is the target vacuum level.

[0075] In some embodiments, updating the position of each particle based on the fitness value includes steps S20131-S20133: S20131. Based on the fitness value, divide the particles in the particle swarm into multiple subgroups; Specifically, particles with different fitness values ​​represent different levels of superiority or inferiority in their positions. For example, if the optimization target is the difference between the second vacuum level and the target vacuum level, as well as the energy loss value of the vacuum pump, then particles with smaller fitness values ​​have better positions (the preset operating parameters are better).

[0076] S20132. For each of the particles, determine the first reference particle as the learning object of the particle from the learning range corresponding to the subgroup to which the particle belongs, and obtain the historical best position of the first reference particle, wherein the historical best position is the position corresponding to the minimum fitness value during the particle position update process. Specifically, by dividing the particles into subgroups based on their fitness values, each subgroup will have a different level of excellence. Therefore, a learning range can be determined for each subgroup; for example, the optimal subgroup's learning range could be itself. For each particle, a first reference particle can be determined from the learning range of its respective subgroup, and learning can be performed based on the first reference particle's historical best position (the position corresponding to the minimum fitness value during position updates).

[0077] S20133. Update the position of the particle based on the historical best position of the first reference particle corresponding to the particle.

[0078] Specifically, during the position update process of each particle, it learns from and approaches the historical best position of its corresponding first reference particle.

[0079] In some embodiments, the particle swarm includes a first subgroup, a second subgroup, and a third subgroup, wherein the fitness value of each particle in the first subgroup is less than the fitness value of each particle in the second subgroup, and the fitness value of each particle in the second subgroup is less than the fitness value of each particle in the third subgroup; for a particle in the first subgroup, its corresponding learning range is the first subgroup, and its corresponding first reference particle includes the particle itself and randomly selected particles from the first subgroup; for a particle in the second subgroup, its corresponding learning range is the first subgroup, the second subgroup, and the third subgroup, and its first reference particle includes particles randomly selected from the first subgroup and the second subgroup, and particles determined from the particle swarm that correspond to the minimum fitness value during the position update process; for a particle in the third subgroup, its corresponding learning range is the first subgroup, the second subgroup, and the third subgroup, and its corresponding first reference particle includes the particle itself and particles determined from the particle swarm that correspond to the minimum fitness value during the position update process.

[0080] Specifically, if the optimization target is the difference between the second vacuum level and the target vacuum level, and the energy loss of the vacuum pump, the particles can be sorted in ascending order of fitness value. Based on the sorting results, the particle swarm can be divided into a first subgroup, a second subgroup, and a third subgroup according to a preset ratio (e.g., 2:3:5). After the division, the first subgroup is the elite population, the second subgroup is the ordinary population, and the third subgroup is the poor population. Corresponding tasks can be assigned to the particles in each subgroup. For example, the tasks corresponding to the first subgroup, the second subgroup, and the third subgroup are development, balancing, and exploration, respectively. Thus, the learning scope of each subgroup is also different.

[0081] In this embodiment, the learning range of particles in the first subgroup can be the first subgroup itself, and the learning object can be the particle itself or particles randomly selected from the first subgroup. Their positions can be updated according to a first formula and a second formula. The first formula is:

[0082] in, It is the position of the d-th preset running parameter of the i-th particle in the first subgroup at time t+1. It is the position of the d-th preset running parameter of the i-th particle in the first subgroup at time t. It is the velocity of the d-th preset operating parameter of the i-th particle in the first subgroup at time t+1.

[0083] The second formula is:

[0084] in, It is the velocity of the d-th preset operating parameter of the i-th particle in the first subgroup at time t. It is the inertial weight of the velocities of particles in the first subgroup. It is the first acceleration coefficient. It is the first random number in the range [0,1] corresponding to the d-th preset running parameter. It represents the historical optimal position of the d-th preset operating parameter of the i-th particle in the first subgroup within the first t time steps. It is the position of the d-th preset running parameter of the i-th particle in the first subgroup at time t. It is the second acceleration coefficient. It is the second random number in the range [0,1] corresponding to the d-th preset running parameter. It is the historical optimal position of the d-th preset running parameter of a particle randomly selected from the first subgroup within the first t time steps. It is the velocity of the d-th preset operating parameter of the i-th particle in the first subgroup at time t+1.

[0085] The learning scope for particles in the second subgroup can be the entire particle swarm (including the first, second, and third subgroups). The learning objects include particles randomly selected from the first and second subgroups, as well as particles determined from the particle swarm that correspond to the minimum fitness value during the position update process (i.e., the globally optimal historical position). The positions of particles in the second subgroup can be updated according to the first and third formulas. The third formula is:

[0086] in, It is the velocity of the d-th preset operating parameter of the i-th particle in the second subgroup at time t. It is the inertial weight of the velocities of particles in the second subgroup. It is the first acceleration coefficient. It is the third random number in the range [0,1] corresponding to the d-th preset running parameter. It represents the historical optimal position of the d-th preset operating parameter of a particle randomly selected from the first and second subgroups within the first t time steps. It is the position of the d-th preset running parameter of the i-th particle in the second subgroup at time t. It is the second acceleration coefficient. It is the fourth random number in the range [0,1] corresponding to the d-th preset running parameter. It represents the globally historical best position of the d-th pre-defined operating parameter in the entire particle swarm within the first t time steps. It is the velocity of the d-th preset operating parameter of the i-th particle in the second subgroup at time t+1.

[0087] The learning scope for particles in the third subgroup can be the entire particle swarm (including the first, second, and third subgroups). The learning objects include the particle itself, as well as the particles with the minimum fitness value determined from the particle swarm during the position update process (i.e., the globally optimal historical position). The positions of particles in the third subgroup can be updated according to the first and fourth formulas. The fourth formula is:

[0088] in, It is the velocity of the d-th preset operating parameter of the i-th particle in the third subgroup at time t. It is the inertial weight of the velocity of particles in the third subgroup. It is the first acceleration coefficient. It is the fifth random number in the range [0,1] corresponding to the d-th preset running parameter. It represents the historical optimal position of the d-th preset operating parameter of the i-th particle in the third subgroup at the previous t-th time step. It is the position of the d-th preset running parameter of the i-th particle in the third subgroup at time t. It is the second acceleration coefficient. It is the sixth random number in the range [0,1] corresponding to the d-th preset running parameter. It represents the globally historical best position of the d-th pre-defined operating parameter in the entire particle swarm within the first t time steps. It is the velocity of the d-th preset operating parameter of the i-th particle in the third subgroup at time t+1.

[0089] In this embodiment, during the process of updating and iterating the particle positions, the proportion of each subgroup can be dynamically adjusted according to the progress of the iteration. For example, in the later stage of the iteration, the proportion of the elite population can be increased and the proportion of the poor population can be reduced to avoid excessive consumption of computing resources in non-high-quality areas.

[0090] In this embodiment, if the parameter values ​​of the particle's velocity and position exceed the parameter boundaries, the parameter values ​​of velocity and position need to be constrained. If the parameter value of velocity is greater than the maximum velocity value, the maximum velocity value is used as the velocity parameter value; if the parameter value of velocity is less than the minimum velocity value, the minimum velocity value is used as the velocity parameter value; if the parameter value of position is greater than the maximum position value, the maximum position value is used as the position parameter value; if the parameter value of position is less than the minimum position value, the minimum position value is used as the position parameter value.

[0091] The calculation formula for velocity constraint processing is as follows:

[0092] in, Let be the velocity of the i-th particle in the particle swarm at time t+1, given the j-th preset operating parameter. The maximum speed of the j-th preset running parameter. The minimum speed value for the j-th preset running parameter.

[0093] The calculation formula for position constraint processing is as follows:

[0094] in, Let j be the position of the i-th particle in the particle swarm at time t+1, based on the j-th preset operating parameter. The position of the j-th preset running parameter is the maximum value. The minimum value of the position of the j-th preset running parameter.

[0095] In some embodiments, the method further includes steps S2001-S2003: S2001. Store the historical best position and corresponding fitness value of each particle in the particle swarm into the database; Specifically, the historical best position of each particle and its corresponding fitness value are stored in a preset database. When the storage exceeds the preset capacity (e.g., N), the particles are sorted in ascending order of fitness value, and the top N historical best positions and their corresponding fitness values ​​are retained to ensure the high quality and lightweight nature of storage resources.

[0096] S2002. When a target particle that meets the preset stagnation condition appears, at least two historical best positions are selected from the database, wherein the historical best position with a smaller fitness value has a higher probability of being selected. Specifically, the preset stagnation condition can be that the number of iterations in which the fitness value does not decrease (i.e., the fitness value does not improve) exceeds a preset number, and the formula for calculating the probability of being selected can be the fifth formula, which is:

[0097] in, Let be the probability that the i-th particle in the database is selected. Let be the fitness value corresponding to the historical best position of the i-th particle in the database. To adjust the parameter (value is 1E) -10 This is to prevent the fitness value of a particle from reaching 0.

[0098] S2003. Update the position of the target particle based on the selected historical best position.

[0099] If two historical best positions are selected, the position of the target particle can be updated using the sixth formula, which is:

[0100] in, Let i be the updated position of the i-th target particle. The first historical best position selected from the database. The second historical best position is selected from the database, and k is the weight parameter.

[0101] In some embodiments, the method further includes steps S20001-S20004: S20001. For each of the particles, a first position difference is obtained based on the historical best position of the particle group and the historical best position of the particle. S20002. Based on the historical best positions of at least two second reference particles randomly selected from the particle swarm, a second position difference is obtained; S20003. Based on the particle's historical best position, the first position difference, and the second position difference, generate a candidate historical best position for the particle; Traditional particle optimization algorithms are prone to premature convergence and lack global search capabilities (finding the optimal parameter values ​​of preset running parameters) during optimization. To solve this problem, this embodiment uses steps S2001-S2004 to update the historical optimal position of each particle.

[0102] The seventh formula can be used to calculate the candidate historical optimal position. The seventh formula is:

[0103] in, It is the candidate historical best position of the i-th particle in the particle swarm. It is the historical best position of the i-th particle in the particle swarm. It is the historical best position of the particle swarm. It is the historical best position of the first second reference particle randomly selected from the particle swarm. It is the historical best position of the second reference particle randomly selected from the particle swarm. It's the difference in the first position. B is the second position difference, and B is the scaling factor (used to control the scaling degree of the first and second position differences).

[0104] S20004. Based on the candidate historical optimal position and the historical optimal position of the particle, the optimized historical optimal position of the particle is obtained.

[0105] In some embodiments, the candidate historical optimal position corresponds to a first parameter value corresponding to multiple preset operating parameters, and the historical optimal position of the particle corresponds to a second parameter value corresponding to multiple preset operating parameters; the step of obtaining the optimized historical optimal position of the particle based on the candidate historical optimal position and the historical optimal position of the particle includes steps S200041-S200042: S200041. Determine the target preset operating parameter from the multiple preset operating parameters; S200042. Modify the parameter value of the target preset running parameter corresponding to the historical best position of the particle from the second parameter value to the first parameter value to obtain the optimized historical best position of the particle.

[0106] Specifically, target operating parameters can be pre-configured, for example, setting the second preset operating parameter as the target operating parameter; or a corresponding random number can be set for each preset operating parameter, and when the random number of a certain preset operating parameter is less than a numerical threshold (for example, the numerical threshold can be a cross factor, and the value of the cross factor is adaptively changing), the preset operating parameter is used as the target preset operating parameter.

[0107] The optimized formula for calculating the historical best position is:

[0108] in, It is the optimized historical best position of the i-th particle in the particle swarm. It represents the candidate historical optimal position of the i-th particle in the particle swarm, and rand(0,1) is a random number corresponding to each preset running parameter. j For the j-th preset operating parameter of the i-th particle in the particle swarm, j rand These are pre-configured target runtime parameters. It is the historical best position of the i-th particle in the particle swarm.

[0109] In some embodiments, after obtaining the optimized historical optimal position of the particle, the method further includes steps S200043-S200045: S200043. For each particle, determine the first fitness value corresponding to the historical best position of the particle, and determine the second fitness value corresponding to the optimized historical best position of the particle. S200044. If the first fitness value is less than or equal to the second fitness value, the historical best position of the particle shall be taken as the updated historical best position of the particle. S200045. If the first fitness value is greater than the second fitness value, the optimized historical best position of the particle shall be taken as the updated historical best position of the particle.

[0110] Specifically, the formula for calculating the updated historical best position is as follows:

[0111] in, It is the historical best position of the i-th particle in the particle swarm after the update. It is the historical best position of the i-th particle in the particle swarm after optimization. It is the historical best position of the i-th particle in the particle swarm. It is the first fitness corresponding to the historical best position of the i-th particle in the particle swarm after optimization. It is the second fitness corresponding to the historical best position of the i-th particle in the particle swarm after optimization.

[0112] In this embodiment, an improved particle swarm optimization algorithm is used to optimize preset operating parameters such as the motor speed and cooling medium flow rate of the vacuum pump, with the difference between the second vacuum level and the target vacuum level, and the energy loss of the vacuum pump as the optimization target. This yields target parameter values ​​for the preset operating parameters, improving the operating efficiency of the vacuum pump, reducing energy consumption, verifying the service life of the vacuum pump, and laying the foundation for improving the accuracy and reliability of fault prediction. Based on the target parameter values, the vacuum pump is operated, and a hybrid model constructed by the mechanism model and the machine learning model is used to predict the faults of the vacuum pump. This overcomes the limitations of single-parameter analysis in the prior art and provides more comprehensive and accurate data support for fault prediction and operation optimization.

[0113] like Figure 3 The diagram shown is a flowchart illustrating the process of determining the target parameter values ​​of the preset operating parameters of the vacuum pump provided in some embodiments of this application. The determination process includes steps S301-S321: S301, Initialization parameters: number of particles in the particle swarm N, number of preset running parameters D, maximum number of iterations maxFES, acceleration coefficients c1 and c2, inertia weight w; S302. Initialize the particle's velocity and position; Among them, the particle corresponds to the preset operating parameters, the particle position indicates the parameter value of the preset operating parameters, and the particle velocity indicates the iteration speed of the particle's preset operating parameter position.

[0114] S303. Calculate the fitness value and update the historical best position P of each particle. i The global optimal position P of the particle swarm g ; S304. Divide the population according to the fitness values; S305, start traversing from the first particle; S306. Determine whether the sorting number of the particles is less than or equal to 0.2*N; Specifically, the particles can be sorted in ascending order of fitness value to obtain the sorting number of each particle. Here, 0.2 is the proportion of particles in the elite group, and a particle's sorting number less than or equal to 0.2*N means that the particle's sorting number is in the top 20.

[0115] S307. When the particle's sorting number is less than or equal to 0.2*N, update the particle's velocity V according to the second formula. i ; S308. If the sorting number of a particle is greater than 0.2*N, determine whether the sorting number of the particle is less than or equal to 0.5*N. Among them, the sorting number of the particles is greater than 0.2*N and less than or equal to 0.5*N, which means that the sorting number of the particles is between 20% and 50%.

[0116] S309. When the particle's sorting number is less than or equal to 0.5*N, update the particle's velocity V according to the third formula. i ; S310. When the particle's sorting number is greater than 0.5*N, update the particle's velocity V according to the fourth formula. i ; If the particle's sort number is greater than 0.5*N, it means that the particle's sort number is in the last 50%.

[0117] S311. Update particle position X according to the first formula. i And calculate the fitness value based on the fitness function; S312, Update the historical best position P of each particle. i The global optimal position P of the particle swarm g ; S313. Determine whether all particles in the particle swarm have been traversed. If not all particles in the particle swarm have been traversed, continue with steps S306-S313.

[0118] S314. If all the particles in the particle swarm have been traversed, and a new round of traversal is performed, is the fitness value of the traversed particles improved? S315. Determine the number of times S indicates that the particle's fitness value has not improved. i Is it less than the preset number T? S316. The number of times S the particle's fitness value is not improved. i If the number of iterations is greater than or equal to the preset number T, then the position of the stopped particle is replaced according to the sixth formula; S317. The number of times S the particle's fitness value is not improved. i If the number of iterations is less than T, or if the position of the stopped particle is replaced according to the sixth formula, then the individual's historical best information is exchanged according to the seventh formula. S318. Determine whether all particles in the particle swarm have been traversed. If the traversal is not complete, repeat steps S314-S318.

[0119] S319. If all particles in the particle swarm have been traversed, record the global optimal position P. g ; S320. Determine if the number of iterations is less than the maximum number of iterations (maxFES). If the number of iterations is less than the maximum number of iterations maxFES, then repeat steps S304-S320. S321. If the number of iterations is greater than or equal to the maximum number of iterations maxFES, then output the optimal motor speed and cooling medium flow rate of the vacuum pump.

[0120] like Figure 4 The diagram shown is another flowchart illustrating a vacuum pump failure prediction method provided in some embodiments of this application. This vacuum pump failure prediction method includes steps S41-S44: S41. The motor speed and cooling medium flow rate of the vacuum pump are set according to the improved particle swarm optimization algorithm. S42. Collect on-site operation and fault data of vacuum pumps, and train them using a BP neural network; S43. The nearest motor speed and cooling medium flow rate of the output vacuum pump are used as inputs to the vacuum pump using process parameters optimized by the algorithm. Field data of the vacuum pump are collected and fault prediction is performed. S44. Output fault prediction results.

[0121] like Figure 5The diagram shown is another schematic flowchart of a vacuum pump fault prediction method provided in some embodiments of this application. This vacuum pump fault prediction method includes steps S501-S512: S501. Input the initialization parameters of the improved particle swarm optimization algorithm: the number of particles N in the particle swarm, the number of preset running parameters D, and the maximum number of iterations maxFES. S502, Input relevant parameters for the vacuum pump; S503, Initialize motor speed and cooling medium flow rate; S504. Divide the population according to the particle fitness value; S505, Update the particle's current velocity and position; S506, obtain the new motor speed and cooling medium flow rate; S507. Calculate the fitness function value with a penalty term; S508. Determine whether the number of iterations has reached the maximum number of iterations, maxFES. S509. If the number of iterations reaches the maximum number of iterations maxFES, collect historical data and use a BP neural network for model training. If the number of iterations has not reached the maximum number of iterations maxFES, repeat steps S504-S509.

[0122] S510. Input the motor speed and cooling medium flow rate obtained from the improved particle swarm optimization algorithm into the vacuum pump parameter settings; S511. Use the trained BP neural network model for fault prediction. S512, Output fault prediction results.

[0123] like Figure 6 The diagram shown is a structural schematic of a vacuum pump fault prediction device 6 provided in some embodiments of this application. The vacuum pump fault prediction device 6 includes: The data acquisition module 61 is used to acquire the operating data of the vacuum pump; The first prediction module 62 is used to input the operating data into the mechanism model in the hybrid model to obtain the first vacuum degree of the vacuum pump. The mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump. The second prediction module 63 is used to input the operating data and the first vacuum degree into the machine learning model in the hybrid model to obtain the failure prediction result of the vacuum pump.

[0124] In some embodiments, the vacuum pump failure prediction device 6 further includes a training module (not shown), which is used for: The historical operating data of the sample vacuum pump is divided into a training set, a validation set, and a test set. The historical operating data has fault labels. The proportion of the same fault label is the same in the historical operating data, the training set, the validation set, and the test set. The machine learning model is iteratively trained using the training set, and the performance of the machine learning model is verified using the validation set during the iterative training process. The hyperparameters of the machine learning model are adjusted based on the verification results to obtain an optimized machine learning model. The optimized machine learning model is evaluated using the test set. If the performance evaluation is passed, the optimized machine learning model is determined as the successfully trained machine learning model.

[0125] In some embodiments, the training module is further configured to: During iterative training, the first performance metric parameters of the machine learning model trained based on the training set are obtained. Obtain the second performance metric parameters of the machine learning model adjusted based on the validation set; If the difference between the first performance index parameter and the second performance index parameter exceeds a preset threshold, the hyperparameters of the machine learning model are adjusted. The machine learning model is then trained based on the adjusted hyperparameters.

[0126] In some embodiments, the fault prediction result includes a state type, which includes a normal state, a potential fault state, and a fault state; when the state type is a potential fault state or a fault state, the fault prediction result also includes a corresponding fault type, which includes at least one of mechanical fault, electrical fault, and performance fault.

[0127] In some embodiments, the vacuum pump failure prediction device 6 further includes an optimization module (not shown), which, before collecting the vacuum pump's operating data, is used to: By optimizing the fitness function, the target parameter values ​​of the preset operating parameters of the vacuum pump are determined. The fitness function represents the mapping relationship between the parameter values ​​corresponding to the preset operating parameters and the fitness value corresponding to the vacuum pump. The vacuum pump is adjusted based on the target parameter value.

[0128] The optimization module is also used for: Based on the position of each particle in the particle swarm, the second vacuum degree of each particle is determined using the aforementioned mechanism model. The particle corresponds to the preset operating parameters of the vacuum pump, and the position of the particle indicates the parameter value of the preset operating parameters. Based on the second vacuum degree, the target vacuum degree, and the fitness function, the fitness value of each particle is determined. The fitness value output by the fitness function is positively correlated with the difference between the second vacuum degree and the target vacuum degree. The target vacuum degree is the vacuum degree of the vacuum pump under ideal vacuum conditions. Update the position of each particle based on the fitness value; Based on the updated positions of each particle, the target parameter value of the preset operating parameters is determined.

[0129] The fitness value of the fitness function is also positively correlated with the energy loss value of the vacuum pump. Before determining the fitness value of each particle based on the second vacuum level, the target vacuum level of the vacuum pump, and the fitness function, the optimization module is further configured to: For each of the aforementioned particles, determine the energy loss value of the vacuum pump at the parameter value indicated by the position of the particle; The fitness value of each particle is determined based on the energy loss value, the second vacuum degree, the target vacuum degree, and the fitness function.

[0130] In some embodiments, the optimization module is further configured to: Based on the fitness value, the particles in the particle swarm are divided into multiple subgroups; For each particle, a first reference particle is determined from the learning range corresponding to the subgroup to which the particle belongs, and the historical optimal position of the first reference particle is obtained. The historical optimal position is the position corresponding to the minimum fitness value during the particle position update process. The position of the particle is updated based on the historical best position of the first reference particle corresponding to the particle.

[0131] In some embodiments, the particle swarm includes a first subgroup, a second subgroup, and a third subgroup, wherein the fitness value of each particle in the first subgroup is less than the fitness value of each particle in the second subgroup, and the fitness value of each particle in the second subgroup is less than the fitness value of each particle in the third subgroup. For a particle in the first subgroup, its corresponding learning range is the first subgroup, and its corresponding first reference particle includes the particle itself and a randomly selected particle in the first subgroup. For particles in the second subgroup, the corresponding learning range is the first subgroup, the second subgroup, and the third subgroup. The first reference particle includes particles randomly selected from the first subgroup and the second subgroup, as well as particles determined from the particle group that correspond to the minimum fitness value during the position update process. For particles in the third subgroup, the corresponding learning range is the first subgroup, the second subgroup, and the third subgroup. The corresponding first reference particle includes the particle itself and the particle with the minimum fitness value determined from the particle group during the position update process.

[0132] In some embodiments, the optimization module is further configured to: Store the historical best position and corresponding fitness value of each particle in the particle swarm in the database; When a target particle that meets the preset stagnation conditions is present, at least two historical best positions are selected from the database, wherein the historical best position with a smaller fitness value has a higher probability of being selected. The position of the target particle is updated based on the selected historical best position.

[0133] In some embodiments, the optimization module is further configured to: For each of the aforementioned particles, a first position difference is obtained based on the historical best position of the particle swarm and the historical best position of the individual particles; The second position difference is obtained based on the historical best positions of at least two second reference particles randomly selected from the particle swarm; Based on the particle's historical best position, the first position difference, and the second position difference, a candidate historical best position for the particle is generated. Based on the candidate historical best position and the historical best position of the particle, the optimized historical best position of the particle is obtained.

[0134] In some embodiments, the candidate historical optimal position corresponds to a first parameter value corresponding to multiple preset operating parameters, and the historical optimal position of the particle corresponds to a second parameter value corresponding to multiple preset operating parameters; the optimization module is further configured to: The target preset operating parameter is determined from the various preset operating parameters; The parameter value of the target preset running parameter corresponding to the historical best position of the particle is modified from the second parameter value to the first parameter value to obtain the optimized historical best position of the particle.

[0135] In some embodiments, after obtaining the optimized historical best position of the particle, the optimization module is further configured to: For each particle, determine the first fitness value corresponding to the particle's historical best position, and determine the second fitness value corresponding to the particle's optimized historical best position. If the first fitness value is less than or equal to the second fitness value, the historical best position of the particle is taken as the updated historical best position of the particle. If the first fitness value is greater than the second fitness value, the optimized historical best position of the particle is taken as the updated historical best position of the particle.

[0136] The vacuum pump fault prediction device 6 provided in this application provides a real and direct data foundation for fault prediction by collecting actual operating data of the vacuum pump. The mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump, which can theoretically reflect the relationship between operating data and vacuum degree. The machine learning model comprehensively utilizes the operating data and the first vacuum degree output by the mechanism model to perform fault analysis on the vacuum pump. This not only gives full play to the advantages of the mechanism model in reflecting physical characteristics, but also utilizes the ability of the machine learning model to identify faults from the data, thereby improving the accuracy and reliability of fault prediction results.

[0137] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0138] In one embodiment, the internal structure diagram of the electronic device can be as follows: Figure 7 As shown, the electronic device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a vacuum pump fault prediction method. The display unit of the electronic device is used to form a visually visible image. It can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the electronic device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the electronic device, or external keyboards, touchpads, or mice, etc.

[0139] Those skilled in the art will understand that Figure 7The structure shown is only a block diagram of a part of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0140] Based on the same inventive concept, embodiments of this application also provide a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0141] Since the computer program stored in the computer-readable storage medium can execute any of the vacuum pump failure prediction methods provided in the embodiments of this application, the beneficial effects that any of the vacuum pump failure prediction methods provided in the embodiments of this application can achieve can be realized, as detailed in the preceding embodiments, and will not be repeated here.

[0142] Based on the same inventive concept, embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the methods provided in the various optional implementations of the above embodiments.

Claims

1. A method for predicting vacuum pump failures, characterized in that, Includes the following steps: Collect operating data of the vacuum pump; The operational data is input into the mechanistic model in the hybrid model to obtain the first vacuum degree of the vacuum pump. The mechanistic model is constructed based on the working principle and physical characteristics of the vacuum pump. The operating data and the first vacuum level are input into the machine learning model in the hybrid model to obtain the failure prediction result of the vacuum pump.

2. The method according to claim 1, characterized in that, The training process of the machine learning model includes: The historical operating data of the sample vacuum pump is divided into a training set, a validation set, and a test set. The historical operating data has fault labels. The proportion of the same fault label is the same in the historical operating data, the training set, the validation set, and the test set. The machine learning model is iteratively trained using the training set, and the performance of the machine learning model is verified using the validation set during the iterative training process. The hyperparameters of the machine learning model are adjusted based on the verification results to obtain an optimized machine learning model. The optimized machine learning model is evaluated using the test set. If the performance evaluation is passed, the optimized machine learning model is determined as the successfully trained machine learning model.

3. The method according to claim 2, characterized in that, The adjustment of the hyperparameters of the machine learning model based on the validation results includes: During iterative training, the first performance metric parameters of the machine learning model trained based on the training set are obtained. Obtain the second performance metric parameters of the machine learning model adjusted based on the validation set; If the difference between the first performance index parameter and the second performance index parameter exceeds a preset threshold, the hyperparameters of the machine learning model are adjusted. The machine learning model is then trained based on the adjusted hyperparameters.

4. The method according to claim 1, characterized in that, The fault prediction result includes a state type, which includes a normal state, a potential fault state, and a fault state. When the state type is a potential fault state or a fault state, the fault prediction result also includes a corresponding fault type, which includes at least one of mechanical fault, electrical fault, and performance fault.

5. The method according to any one of claims 1-4, characterized in that, Before acquiring the operating data of the vacuum pump, the method further includes: By optimizing the fitness function, the target parameter values ​​of the preset operating parameters of the vacuum pump are determined. The fitness function represents the mapping relationship between the parameter values ​​corresponding to the preset operating parameters and the fitness value corresponding to the vacuum pump. The vacuum pump is adjusted based on the target parameter value.

6. The method according to claim 5, characterized in that, The step of determining the target parameter values ​​of the preset operating parameters of the vacuum pump by optimizing the fitness function includes: Based on the position of each particle in the particle swarm, the second vacuum degree of each particle is determined using the aforementioned mechanism model. The particle corresponds to the preset operating parameters of the vacuum pump, and the position of the particle indicates the parameter value of the preset operating parameters. Based on the second vacuum degree, the target vacuum degree, and the fitness function, the fitness value of each particle is determined. The fitness value output by the fitness function is positively correlated with the difference between the second vacuum degree and the target vacuum degree. The target vacuum degree is the vacuum degree of the vacuum pump under ideal vacuum conditions. Update the position of each particle based on the fitness value; Based on the updated positions of each particle, the target parameter value of the preset operating parameters is determined.

7. The method according to claim 6, characterized in that, The fitness value of the fitness function is also positively correlated with the energy loss value of the vacuum pump. Before determining the fitness value of each particle based on the second vacuum level, the target vacuum level, and the fitness function, the method further includes: For each of the aforementioned particles, determine the energy loss value of the vacuum pump at the parameter value indicated by the position of the particle; The determination of the fitness value of each particle based on the second vacuum degree, the target vacuum degree, and the fitness function includes: The fitness value of each particle is determined based on the energy loss value, the second vacuum degree, the target vacuum degree, and the fitness function.

8. The method according to claim 6, characterized in that, Updating the position of each particle based on the fitness value includes: Based on the fitness value, the particles in the particle swarm are divided into multiple subgroups; For each particle, a first reference particle is determined from the learning range corresponding to the subgroup to which the particle belongs, and the historical optimal position of the first reference particle is obtained. The historical optimal position is the position corresponding to the minimum fitness value during the particle position update process. The position of the particle is updated based on the historical best position of the first reference particle corresponding to the particle.

9. The method according to claim 8, characterized in that, The particle swarm includes a first subgroup, a second subgroup, and a third subgroup, wherein the fitness value of each particle in the first subgroup is less than the fitness value of each particle in the second subgroup, and the fitness value of each particle in the second subgroup is less than the fitness value of each particle in the third subgroup. For a particle in the first subgroup, its corresponding learning range is the first subgroup, and its corresponding first reference particle includes the particle itself and a randomly selected particle in the first subgroup. For particles in the second subgroup, the corresponding learning range is the first subgroup, the second subgroup, and the third subgroup. The first reference particle includes particles randomly selected from the first subgroup and the second subgroup, as well as particles determined from the particle group that correspond to the minimum fitness value during the position update process. For particles in the third subgroup, the corresponding learning range is the first subgroup, the second subgroup, and the third subgroup. The corresponding first reference particle includes the particle itself and the particle with the minimum fitness value determined from the particle group during the position update process.

10. The method according to claim 8, characterized in that, The method further includes: Store the historical best position and corresponding fitness value of each particle in the particle swarm in the database; When a target particle that meets the preset stagnation conditions is present, at least two historical best positions are selected from the database, wherein the historical best position with a smaller fitness value has a higher probability of being selected. The position of the target particle is updated based on the selected historical best position.

11. The method according to claim 8, characterized in that, The method further includes: For each of the aforementioned particles, a first position difference is obtained based on the historical best position of the particle swarm and the historical best position of the individual particles; The second position difference is obtained based on the historical best positions of at least two second reference particles randomly selected from the particle swarm; Based on the particle's historical best position, the first position difference, and the second position difference, a candidate historical best position for the particle is generated. Based on the candidate historical best position and the historical best position of the particle, the optimized historical best position of the particle is obtained.

12. The method according to claim 11, characterized in that, The candidate historical optimal position corresponds to a first parameter value of multiple preset operating parameters, and the historical optimal position of the particle corresponds to a second parameter value of multiple preset operating parameters. The step of obtaining the optimized historical optimal position of the particle based on the candidate historical optimal position and the historical optimal position of the particle includes: The target preset operating parameter is determined from the various preset operating parameters; The parameter value of the target preset running parameter corresponding to the historical best position of the particle is modified from the second parameter value to the first parameter value to obtain the optimized historical best position of the particle.

13. The method according to claim 12, characterized in that, After obtaining the optimized historical best position of the particle, the method further includes: For each particle, determine the first fitness value corresponding to the particle's historical best position, and determine the second fitness value corresponding to the particle's optimized historical best position. If the first fitness value is less than or equal to the second fitness value, the historical best position of the particle is taken as the updated historical best position of the particle. If the first fitness value is greater than the second fitness value, the optimized historical best position of the particle is taken as the updated historical best position of the particle.

14. A vacuum pump fault prediction device, characterized in that, The device includes: The data acquisition module is used to collect the operating data of the vacuum pump; The first prediction module is used to input the operating data into the mechanism model in the hybrid model to obtain the first vacuum degree of the vacuum pump. The mechanism model is constructed based on the working principle and physical characteristics of the vacuum pump. The second prediction module is used to input the operating data and the first vacuum degree into the machine learning model in the hybrid model to obtain the failure prediction result of the vacuum pump.

Images