Vacuum furnace equipment fault intelligent diagnosis method based on AI

Abstract

The invention discloses a vacuum furnace equipment fault intelligent diagnosis method based on AI, and belongs to the technical field of equipment fault diagnosis. According to the method, fault diagnosis data of the vacuum furnace are obtained through a sensor, fault diagnosis features are obtained through preprocessing and feature engineering processing, a vacuum furnace AI equipment fault diagnosis model is constructed accordingly, and the fault diagnosis features are processed through the vacuum furnace AI equipment fault diagnosis model to obtain a fault type. And performing graph traversal search recommendation on the fault types through a fault knowledge graph constructed based on historical maintenance records to obtain fault causes corresponding to the fault types. According to the method, AI and knowledge graph technologies are fused, full-process intelligent diagnosis from data acquisition to fault reason positioning is realized, a scientific basis is provided for vacuum furnace equipment maintenance, the problems of low efficiency and poor accuracy of a traditional diagnosis method are solved, and the intelligent level and reliability of equipment fault diagnosis are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment fault diagnosis, and in particular to an AI-based vacuum furnace equipment fault intelligent diagnosis method. BACKGROUND

[0002] As a key equipment in the field of heat treatment and metallurgy, the running stability of the vacuum furnace directly affects the product quality and production efficiency. Traditional vacuum furnace fault diagnosis mainly relies on manual inspection and experience judgment, which has the defects of low diagnosis efficiency, poor accuracy, insufficient data utilization and difficult knowledge inheritance: manual checking of equipment parameters and fault phenomena one by one, it is difficult to quickly locate the root cause, and the diagnosis cycle is long in complex fault scenarios; it is easy to be affected by subjective factors and has insufficient ability to identify early hidden faults, relying on the experience of maintenance personnel; the multi-dimensional data such as temperature, vibration and current collected by the equipment sensors are not effectively integrated and analyzed, and a systematic fault diagnosis system cannot be formed; historical maintenance experience is mainly recorded in text, lacks structured association, is difficult to reuse and iterate, and leads to repeated troubleshooting of similar faults.

[0003] With the development of industrial intelligence, although some schemes try to introduce machine learning technology, they are mostly limited to single-parameter analysis of vibration signals and other single parameters, do not integrate multi-source data to build a comprehensive diagnosis model, and lack structured use of historical maintenance knowledge, so they cannot realize the correlation deduction from fault type to fault cause. Therefore, an intelligent diagnosis method combining AI and knowledge graph technology is urgently needed to solve the problems of low efficiency and poor accuracy of traditional diagnosis and improve the intelligence and reliability of vacuum furnace equipment fault diagnosis. Therefore, an AI-based vacuum furnace equipment fault intelligent diagnosis method is proposed. SUMMARY

[0004] The present application aims to provide an AI-based vacuum furnace equipment fault intelligent diagnosis method to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides the following technical solutions: An AI-based vacuum furnace equipment fault intelligent diagnosis method, comprising the following steps: S1. Acquiring fault diagnosis data of the vacuum furnace through a sensor; S2. Preprocessing the fault diagnosis data, processing the preprocessed fault diagnosis data through feature engineering to obtain fault diagnosis features, and storing the fault diagnosis features in a database; S3. Constructing a vacuum furnace AI equipment fault diagnosis model, inputting the fault diagnosis features into the vacuum furnace AI equipment fault diagnosis model for processing to obtain a fault type; S4. Constructing a fault knowledge graph based on historical maintenance records, searching the fault knowledge graph according to the fault type to obtain a fault cause corresponding to the fault type.

[0006] Preferably, the fault diagnosis data of the vacuum furnace is obtained by the sensor, and the fault diagnosis data of the vacuum furnace includes heating zone temperature, furnace vacuum degree, cooling water inlet and outlet temperature, pump group vibration signal, motor current, process gas flow and cooling water flow.

[0007] Preferably, the method for pre-processing the fault diagnosis data comprises the following steps: The non-vibration data is pre-processed by the steps of: calculating the mean value and the standard deviation of each non-vibration data sequence , determining the data points exceeding the interval as outliers and deleting the outliers, and using the Lagrange interpolation method to restore the missing values of the non-vibration data sequence after the outliers are processed, for a non-vibration data sequence with a length of n, if the data at the kth position is missing, an m-1 order polynomial is constructed by using the adjacent m effective data points , satisfying , so as to estimate the missing value , thereby ensuring the continuity of the data time sequence, and using the Z-score standardization method to eliminate the dimension influence and perform standardization processing on the non-vibration data after the missing values are filled, so that the non-vibration data after the missing values are filled conforms to the standard normal distribution with a mean value of 0 and a standard deviation of 1, and the pre-processed non-vibration data is obtained; The pump group vibration signal is restored by using the wavelet transform reconstruction technology, the pump group vibration signal is decomposed into wavelet coefficients of different frequency bands, the wavelet coefficients of the adjacent effective segments are interpolated after the missing segment, and then the time domain signal is inversely transformed and reconstructed, so that the pump group vibration signal after the missing signal is restored is obtained, and the minimum-maximum normalization method is used for standardization processing, so that the pump group vibration signal after the missing signal is restored is mapped to the interval [0, 1], and the pre-processed pump group vibration signal is obtained, and the pre-processed fault diagnosis data is obtained accordingly; The non-vibration data includes heating zone temperature, furnace vacuum degree, cooling water inlet and outlet temperature, motor current, process gas flow and cooling water flow. Preferably, the method for pre-processing the fault diagnosis data comprises the following steps:

[0008] The pre-processed pump group vibration signal , heating zone temperature and process gas flow are calculated by using the vibration kurtosis formula, temperature gradient difference formula and gas consumption anomaly index formula respectively, so as to obtain vibration kurtosis , temperature gradient difference and gas consumption anomaly index , and the pre-processed pump group vibration signal is processed by using the feature engineering method. The fast Fourier transform is performed to obtain a frequency domain signal , according to which the frequency domain signal The power spectral density is calculated by a periodogram method , and the bearing fault energy ratio is calculated by a bearing fault energy ratio formula ; The preprocessed heating zone temperature and the pump group vibration signal The vibration temperature correlation is calculated by a vibration temperature correlation formula , the preprocessed motor current and the furnace vacuum degree The vacuum current delay is calculated by a vacuum current delay formula , and the preprocessed cooling water inlet and outlet temperature and , and the cooling water flow The cooling efficiency is calculated by a cooling efficiency formula ; The furnace vacuum degree standard deviation and the furnace vacuum degree mean value of the preprocessed furnace vacuum degree , the furnace vacuum degree standard deviation and the furnace vacuum degree mean value The pressure fluctuation coefficient is calculated by a pressure fluctuation coefficient formula ; The preprocessed motor current The wave current effective value and the harmonic current effective value are obtained by fast Fourier transform , the wave current effective value and the harmonic current effective value are brought into a current harmonic distortion rate formula to calculate the current harmonic distortion rate , thereby obtaining the fault diagnosis features The vibration kurtosis formula is: ; wherein, is the number of data points, is the standard deviation of the preprocessed pump group vibration signal , and is the mean value of the preprocessed pump group vibration signal ; The temperature gradient difference formula is: ; wherein, is the heating zone temperature at six consecutive time points of the preprocessed heating zone temperature ; The gas consumption anomaly index formula is: is a preset process gas flow; The bearing fault energy ratio formula is: The vibration temperature correlation formula is: is the mean value of the pretreated pump group vibration signal is the mean value of the pretreated heating zone temperature is the number of data points; The cooling efficiency formula is: The pressure fluctuation coefficient formula is: The current harmonic distortion rate formula is: The vacuum current delay formula is: is the delay; The periodogram method is a method for calculating the power spectral density (PSD).

[0009] Preferably, the method for constructing a vacuum furnace AI equipment fault diagnosis model comprises: Obtaining a plurality of historical fault diagnosis features from a database to form a fault diagnosis feature set, dividing the fault diagnosis feature set into a training set and a test set according to a 7:3 ratio, manually labeling the fault types of the training set and the test set, inputting the training set into a random forest for training, and inputting the test set into the trained random forest to obtain the predicted fault types, counting the number of predicted error samples whose predicted fault types are different from the manually labeled fault types of the test set, if the proportion of the predicted error samples to the sample number of the test set exceeds 2.5%, retraining the random forest, otherwise, obtaining the trained random forest, i.e. the vacuum furnace AI equipment fault diagnosis model, wherein the fault types include bearing fault, heater fault, gas leakage fault, cooling system fault, vacuum pump fault and motor fault.

[0010] Preferably, the method for constructing a fault knowledge graph based on historical maintenance records comprises: ​​​​​​​​​​​​Natural language processing technology is used to extract entities, relations and attributes from historical maintenance records, and a fault knowledge graph is constructed using the graph database Neo4j. Entities are used as nodes in the fault knowledge graph, relations are used as edges, and attributes are used as edge attributes. Entities represent fault types and fault causes, relations represent the association between fault types and fault causes, and attributes represent the frequency of occurrence of fault causes, maintenance time of fault causes, maintenance cost of fault causes, and recurrence rate of fault causes.

[0011] Preferably, the method for obtaining the fault cause corresponding to the fault type is as follows: Based on the fault type, a graph traversal search (breadth-first search or depth-first search) is performed on the fault knowledge graph to obtain all fault knowledge graph nodes corresponding to fault causes that have fault knowledge graph edges corresponding to the fault type. For the fault knowledge graph edge attributes corresponding to all fault cause nodes, the fault cause probability degree of each fault cause node is calculated using the fault cause probability measurement formula. The fault knowledge graph node corresponding to the fault cause with the highest fault cause probability degree is selected, thus obtaining the fault cause corresponding to the fault type. The possible formula for determining the cause of the failure is: ; in, and These are the frequency of occurrence of the cause of the failure, the repair time for the cause of the failure, the repair cost for the cause of the failure, and the recurrence rate of the cause of the failure. and These are the information entropy values ​​for the frequency of occurrence of the cause of the failure, the repair time for the cause of the failure, the repair cost for the cause of the failure, and the recurrence rate of the cause of the failure. Entropy for attribute conflicts; The attribute conflict entropy for: ; in, and These are the average values ​​of the frequency of occurrence of the cause of failure, the repair time of the cause of failure, the repair cost of the cause of failure, and the recurrence rate of the cause of failure. The for: ; in, yes ( For the first The first attribute (The attribute values ​​of each fault cause), where n is the number of fault causes.

[0012] Compared with the prior art, the present application has the following beneficial effects: 1. The present application obtains multi-dimensional fault diagnosis data through sensors, adopts 3 principles, Lagrange interpolation method, wavelet transform reconstruction technology and standardization for pretreatment, and extracts 9-dimensional fault diagnosis features through feature engineering, compared with the limitations of single parameter analysis in the prior art, the present application realizes deep fusion and intelligent processing of multi-source data, can comprehensively reflect the running state of the vacuum furnace equipment from different dimensions, and effectively improves the comprehensiveness and accuracy of fault diagnosis.

[0013] 2. The present application extracts entities, relationships and attributes from historical maintenance records using natural language processing technology, constructs a fault knowledge graph through Neo4j, and designs a fault reason possibility degree calculation formula based on weighted geometric mean and entropy weight correction, compared with the situation that historical maintenance experience is mainly recorded in text and lacks structured association in the prior art, the present application can quickly retrieve and recommend the most possible fault reason in the knowledge graph according to the fault type, simultaneously considers multi-dimensional attributes such as fault reason occurrence frequency and fault reason maintenance time, realizes accurate positioning of the fault reason, provides scientific maintenance reference for maintenance personnel, significantly improves maintenance efficiency, and shortens equipment downtime.

[0014] 3. The present application integrates AI and knowledge graph technology, realizes whole-process intelligent diagnosis from data acquisition, pretreatment, feature engineering, fault diagnosis to fault reason positioning, compared with the traditional diagnosis method relying on manual inspection and experience judgment, the present application effectively solves the problems of low efficiency and poor accuracy of the traditional diagnosis method, reduces the influence of human factors, improves the intelligent level and reliability of equipment fault diagnosis, and provides strong technical support for safe and stable operation of the vacuum furnace equipment. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows, obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0016] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0018] Examples, such as Figure 1 As shown, an AI-based intelligent fault diagnosis method for vacuum furnace equipment includes the following steps: S1. Obtain fault diagnosis data of the vacuum furnace through sensors; S2. Preprocess the fault diagnosis data, and then process the preprocessed fault diagnosis data through feature engineering to obtain fault diagnosis features and store them in the database; S3. Construct a fault diagnosis model for vacuum furnace AI equipment, input fault diagnosis features into the fault diagnosis model for processing, and obtain the fault type; S4. Construct a fault knowledge graph based on historical maintenance records, and retrieve the fault knowledge graph according to the fault type to obtain the fault cause corresponding to the fault type.

[0019] Furthermore, the working principle of the present invention will be illustrated below through embodiments: Take a vacuum furnace of a certain model with a rated temperature of 1200℃ and an ultimate vacuum of 5×10⁻³Pa as an example.

[0020] K-type thermocouples were installed in the heating zone of the vacuum furnace, piezoresistive pressure sensors were installed inside the furnace, PT100 temperature sensors were installed at the cooling water inlet and outlet, acceleration and vibration sensors were installed on the pump bearing housing, current transformers were connected to the motor and heater circuits, and thermal mass flow meters were installed on the process gas pipelines. Fault diagnosis data was collected in real time using sensors at a sampling frequency of 100Hz via a PLC control system and stored in the InfluxDB database. The data collection period was 72 hours of continuous equipment operation, yielding approximately 2.5 × 10⁻⁶ data. 7 Original records of fault diagnosis data.

[0021] The non-vibration data in the fault diagnosis data is preprocessed, taking the heating zone temperature as an example. The mean of the thermocouple temperature sequence is 850℃, and the standard deviation is 15℃. The heating zone temperature less than 805℃ or greater than 895℃ is deleted, and a total of 327 abnormal values are identified. After the abnormal value processing, the 10-minute data of the heating zone temperature is missing. Accordingly, a 4th order polynomial is constructed by using the 5-point Lagrange interpolation method through the adjacent time temperature values T(t-2), T(t-1), T(t+1) and T(t+2), and the missing value is estimated as 848.5℃. At the same time, the heating zone temperature after filling the missing value is standardized by Z-score, and the preprocessed heating zone temperature is obtained. According to the method of obtaining the preprocessed heating zone temperature, the same processing is performed on the remaining non-vibration data. The pump group vibration signal is decomposed by db4 wavelet for 3 layers, and the high frequency coefficients and low frequency coefficients are linearly interpolated, respectively. The correlation of the reconstructed time domain waveform with the original pump group vibration signal reaches 0.92, and the standardization processing is performed by the minimum-maximum normalization method. The vibration amplitude of the pump group vibration signal for missing signal recovery is mapped from [-20g, 15g] to [0, 1], and the preprocessed fault diagnosis data is obtained accordingly.

[0022] The preprocessed fault diagnosis data is processed by feature engineering to obtain fault diagnosis features. Taking the extraction of vibration kurtosis and bearing fault energy ratio as an example; 1024 points of preprocessed pump group vibration signal are taken, and the mean is 0.12, the standard deviation is 0.35, and the kurtosis is 6.8 (kurtosis should be less than 5 in normal case, and greater than 5 indicates that the bearing may have impact fault); after FFT transformation of the preprocessed pump group vibration signal, the power spectral density integral value in the 10Hz bandwidth near the bearing inner ring fault characteristic frequency 120Hz is calculated is 0.78, is 2.31, and accordingly the bearing fault energy ratio is about 0.34 (the bearing fault energy ratio should be less than 0.15 in normal case, and greater than 0.15 indicates that the bearing inner ring is worn).

[0023] 5000 groups of historical fault diagnosis features are extracted from the database, and are divided into 3500 groups of training set and 1500 groups of test set according to 7:3. The fault types are manually labeled, and the RandomForestClassifier of scikit-learn is used. The number of trees n_estimators is set to 100, and the maximum depth max_depth is set to 15. After training, the prediction result of the test set meets not more than 2.5%, and accordingly the vacuum furnace AI equipment fault diagnosis model is obtained. The fault diagnosis features are input into the vacuum furnace AI equipment fault diagnosis model for processing, and the fault type is obtained.

[0024] From 1000 historical maintenance records, extract entities such as: fault type is bearing failure and heater failure; fault reason is bearing wear and heater short circuit; attribute is bearing wear frequency is 0.6, maintenance time is 4 hours, cost is 2000 yuan and recurrence rate is 0.2, create directed edge through Neo4j, such as bearing failure → bearing wear, edge attribute is { :0.6, :0.4, :0.7, :0.2}, wherein the values are normalized to [0, 1], when the vacuum furnace AI equipment fault diagnosis model diagnoses the fault type as bearing failure, search the associated fault reason nodes through BFS, such as bearing wear and bearing lubrication failure, calculate the fault reason possibility; the fault reason possibility of bearing wear is calculated through the fault reason possibility is about 0.44, and the fault reason possibility of bearing lubrication failure is calculated through the corresponding fault reason possibility measurement formula as 0.39, so the bearing wear is recommended as the most possible reason, and the maintenance suggestion is to replace the bearing.

[0025] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; modifying the technical solutions described in the foregoing embodiments, or equivalently replacing part of the technical features, does not make the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. An AI-based intelligent fault diagnosis method for vacuum furnace equipment, characterized in that, Includes the following steps: S1. Obtain fault diagnosis data of the vacuum furnace through sensors; S2. Preprocess the fault diagnosis data, and then process the preprocessed fault diagnosis data through feature engineering to obtain fault diagnosis features and store them in the database; The method for feature engineering processing is as follows: The pre-processed pump vibration signal, heating zone temperature, and process gas flow rate were calculated using the vibration kurtosis formula, temperature gradient difference formula, and gas consumption anomaly index formula, respectively, to obtain the vibration kurtosis. Temperature gradient difference and gas consumption anomaly index Simultaneously, the frequency domain signal is obtained by performing a fast Fourier transform on the preprocessed pump vibration signal. Based on this, the frequency domain signal The power spectral density was calculated using the periodogram method. The bearing failure energy ratio is calculated by substituting it into the bearing failure energy ratio formula. ; The vibration-temperature correlation was calculated using the vibration-temperature correlation formula for the pre-treated heating zone temperature and pump unit vibration signal. Simultaneously, the pre-processed motor current and furnace vacuum level are used to calculate the vacuum current delay using the vacuum current delay formula. The cooling efficiency was calculated using the pretreated inlet and outlet temperatures of the cooling water and the cooling water flow rate, based on the cooling efficiency formula. ; Calculate the standard deviation and mean of the furnace vacuum degree after pretreatment. Then, use the standard deviation and mean of the furnace vacuum degree to calculate the pressure fluctuation coefficient using the pressure fluctuation coefficient formula. ; The effective value of the preprocessed motor current is obtained by fast Fourier transform. and the effective value of each harmonic current The current harmonic distortion rate is obtained by substituting the effective value of the wave current and the effective value of each harmonic current into the formula for current harmonic distortion rate. Based on this, fault diagnosis characteristics are obtained. ; S3. Construct a fault diagnosis model for vacuum furnace AI equipment, input fault diagnosis features into the fault diagnosis model for processing, and obtain the fault type; S4. Construct a fault knowledge graph based on historical maintenance records, and retrieve the fault knowledge graph according to the fault type to obtain the fault cause corresponding to the fault type.

2. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 1, characterized in that, The fault diagnosis data of the vacuum furnace obtained by the sensors includes heating zone temperature, furnace vacuum degree, cooling water inlet and outlet temperatures, pump vibration signal, motor current, process gas flow rate, and cooling water flow rate.

3. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 2, characterized in that, The method for preprocessing fault diagnosis data: For non-vibration data, through 3 The principle is to perform outlier processing, identify and delete outliers, and then fill missing values ​​in the outlier-processed non-vibration data using Lagrange interpolation. At the same time, the non-vibration data after missing value filling is standardized using the Z-score standardization method to obtain preprocessed non-vibration data. The pump vibration signal is recovered by wavelet transform reconstruction technology, and then standardized by the minimum-maximum normalization method to obtain the preprocessed pump vibration signal, and the preprocessed fault diagnosis data is obtained accordingly. The non-vibration data includes heating zone temperature, furnace vacuum, cooling water inlet and outlet temperatures, motor current, process gas flow rate, and cooling water flow rate.

4. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 1, characterized in that, The method for constructing a fault diagnosis model for a vacuum furnace AI device: Several historical fault diagnosis features are obtained from the database to form a fault diagnosis feature set. The fault diagnosis feature set is divided into a training set and a test set in a 7:3 ratio. Fault types are manually labeled on the training set and the test set. The training set is input into a random forest for training, and the test set is input into the trained random forest to obtain the predicted fault types. The number of prediction error samples that are different from the manually labeled fault types in the test set is counted. If the proportion of the number of prediction error samples to the number of samples in the test set exceeds 2.5%, the random forest is retrained. Otherwise, the trained random forest, i.e., the fault diagnosis model of the vacuum furnace AI equipment, is obtained.

5. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 4, characterized in that, The fault types include bearing faults, heater faults, gas leak faults, cooling system faults, vacuum pump faults, and motor faults.

6. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 5, characterized in that, The method for constructing a fault knowledge graph based on historical maintenance records: Natural language processing technology is used to extract entities, relations and attributes from historical maintenance records, and a fault knowledge graph is constructed using the graph database Neo4j. Entities are used as nodes in the fault knowledge graph, relations are used as edges, and attributes are used as edge attributes. Entities represent fault types and fault causes, relations represent the association between fault types and fault causes, and attributes represent the frequency of occurrence of fault causes, maintenance time of fault causes, maintenance cost of fault causes, and recurrence rate of fault causes.

7. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 6, characterized in that, The method for obtaining the fault cause corresponding to the fault type: Based on the fault type, a graph traversal search is performed on the fault knowledge graph to obtain all fault knowledge graph nodes corresponding to fault causes that have fault knowledge graph edges with the fault knowledge graph node corresponding to the fault type. For the fault knowledge graph edge attributes corresponding to all fault cause nodes, the fault cause probability degree of the fault knowledge graph node corresponding to each fault cause is calculated using the fault cause probability measurement formula. The fault knowledge graph node corresponding to the fault cause with the highest fault cause probability degree is selected, thus obtaining the fault cause corresponding to the fault type.

8. The AI-based intelligent fault diagnosis method for vacuum furnace equipment according to claim 7, characterized in that, The graph traversal search is performed using either breadth-first search or depth-first search.

Images