Intelligent operation and maintenance system for low-speed heavy-load multi-working-condition equipment

Abstract

This invention relates to the field of equipment condition monitoring and operation and maintenance technology, specifically disclosing an intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment. The system includes: a data acquisition module that collects equipment operating status data from key components in a sensor array according to preset sampling rules; a data processing module that identifies the operating condition type of the pre-processed equipment operating status data, clusters and extracts multiple operating conditions of the equipment operating status, and pre-trains a first-level fault diagnosis model associated with these operating conditions; a data operation and maintenance analysis module that receives the first fault feature parameters, performs fusion analysis on the first fault feature parameters of various historical operating conditions to obtain a standard health index; and compares the first fault feature parameters with a digital twin model to obtain fault diagnosis results, and makes operation and maintenance judgment decisions based on the fault diagnosis results.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring and operation and maintenance technology, specifically to an intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment. Background Technology

[0002] Low-speed, heavy-duty equipment, such as large mining crushers, port ship unloaders, and metallurgical rolling mills, is core equipment in industrial production. Its operating characteristics include low speed, high load, frequent start-stop cycles, and complex operating conditions. Failures in this type of equipment can lead to significant economic losses and safety accidents.

[0003] For low-speed, heavy-load, multi-condition equipment, existing monitoring systems mainly adopt the following monitoring modes:

[0004] Poor adaptability to operating conditions: Most systems use a single fault diagnosis model, which cannot adapt to the multi-condition operation characteristics of equipment under different loads and speeds, resulting in low diagnostic accuracy.

[0005] Insufficient early warning capability: Traditional vibration analysis is mostly based on threshold alarms, which makes it difficult to identify weak features in the early stages of a fault and thus cannot achieve predictive maintenance.

[0006] Information silos: Multi-source sensor information such as vibration, temperature, and acoustic emission is not effectively integrated, resulting in a single dimension of data analysis and making it difficult to comprehensively assess the health status of equipment.

[0007] Model rigidity: Once deployed, the diagnostic model remains fixed and cannot be optimized using newly generated fault data, resulting in a decline in diagnostic capabilities over time.

[0008] Therefore, there is an urgent need for an operation and maintenance system that can intelligently identify operating conditions, integrate multi-source information, and has self-learning capabilities, in order to achieve accurate diagnosis and predictive maintenance of low-speed, heavy-load equipment. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent operation and maintenance system for low-speed heavy-load equipment that can adapt to multiple operating conditions and achieve accurate diagnosis and prediction of early faults.

[0010] The objective of this invention can be achieved through the following technical solutions:

[0011] A smart operation and maintenance system for low-speed, heavy-load, multi-condition equipment constructs a closed-loop smart operation and maintenance ecosystem through a technical path of condition perception, condition classification, condition diagnosis, digital twin verification, and model self-evolution. Specifically, it includes:

[0012] The equipment-layer sensor array is deployed in key parts of low-speed, heavy-load equipment; the equipment-layer sensor array includes at least two of the following: vibration acceleration sensors, acoustic emission sensors, temperature sensors, and speed sensors;

[0013] The data acquisition module communicates with the sensor array at the device layer and is used to acquire equipment operating status data of key parts in the sensor array according to preset sampling rules; the equipment operating status data includes at least two of the following: vibration signal, acoustic emission signal, temperature signal, and rotational speed signal;

[0014] The data processing module is used to identify the operating condition type of the preprocessed equipment operating status data, cluster and extract multiple operating conditions of the equipment operating status, perform machine learning on each operating condition, pre-train and obtain the first-level fault diagnosis model associated with the operating conditions, and input real-time data into the first-level fault diagnosis model for feature extraction, and determine the first fault feature parameters based on the fault matching degree of the corresponding operating conditions.

[0015] The data operation and maintenance analysis module is used to receive the first fault characteristic parameters, perform fusion analysis on the first fault characteristic parameters under various historical operating conditions to obtain the standard health index of the core components of the equipment; compare the actual monitored first fault characteristic parameters with the simulated fault characteristic parameters of the digital twin model under the fault state to obtain the fault diagnosis results, generate the second fault characteristic parameters, and make operation and maintenance judgment decisions based on the fault diagnosis results.

[0016] Preferably, the preset sampling rules include:

[0017] The sampling frequency is set according to the equipment operation cycle. High-frequency sampling is used during the equipment start-up and shutdown phases, and low-frequency sampling is used during the stable operation phase. When the data acquisition module detects that the signal amplitude exceeds the preset safety threshold, the transient high-frequency sampling mode is automatically triggered.

[0018] Preferably, the method for identifying the operating condition type of the preprocessed equipment operating status data is as follows: a feature extraction method based on multi-sensor data fusion is used to construct a comprehensive operating condition feature vector; and principal component analysis is used to reduce the dimensionality of the comprehensive operating condition feature vector, with the dimensionality-reduced principal components used as the input of the clustering algorithm.

[0019] Preferably, the multiple operating conditions of the cluster extraction device are as follows:

[0020] The optimal number of operating condition clusters for principal components is determined using the silhouette coefficient method. ;

[0021] The fuzzy C-means clustering algorithm was used to calculate the operating status data samples for each device. For each working condition cluster membership degree ∈ And satisfy ;

[0022] The operating condition type of each device's operating status data sample is determined by the operating condition cluster corresponding to its maximum membership degree.

[0023] Preferably, a first-level fault diagnosis model is obtained by pre-training machine learning on each working condition. Specifically, the method is as follows:

[0024] For each operating condition cluster A dedicated deep convolutional neural network model was built and trained separately. The input to the model was the time-frequency diagram of the equipment vibration signal obtained after wavelet transform processing. The output layer of the model used the Softmax function to output the probability distribution of the equipment in various fault states under this operating condition. Historical data and corresponding fault labels under different operating conditions are extracted, and supervised learning algorithms, including support vector machines or deep neural networks, are used for training. The hyperparameters of the model are optimized through grid search and k-fold cross-validation, ultimately generating a cluster of operating conditions. The corresponding first-level fault diagnosis model.

[0025] Preferably, the first fault characteristic parameter is determined as follows:

[0026] Probability distribution of obtaining this real-time data With the preset fault probability threshold vector ,and The matching degree is calculated as follows:

[0027]

[0028] when At that time, it is determined that the equipment is prone to failure, and will cause The fault type that achieves the maximum value and its matching degree are both used as the first fault characteristic parameter. Output; where, This is a real-time fault type; For real-time matching accuracy; For real-time probability distribution; This is the real-time value of the preset fault probability threshold vector.

[0029] Preferably, the first fault characteristic parameters under various historical operating conditions are fused and analyzed to obtain the standard health index of the core components of the equipment. The calculation process is as follows:

[0030] Collect the set of first fault characteristic parameters under various operating conditions over a historical period;

[0031] Feature fusion is performed on the first set of fault characteristic parameters to obtain the standard health index. The weighted calculation formula is used to obtain the following:

[0032]

[0033] in, This represents the total number of fault types. Fault type Average matching degree over a historical period; Fault type The weighting coefficients are determined based on the historical frequency of the fault and the cost of fault repair. Principal component analysis is used to reduce the dimensionality of the first fault characteristic parameters under various historical operating conditions. Then, a weighted average method is used to fuse the scores of each principal component. The weighting coefficients are determined based on the frequency and severity of the fault under each operating condition. Finally, a standard health index between 0 and 1 is obtained, where 1 represents complete health and 0 represents severe fault.

[0034] Preferably, the second fault diagnosis result is presented in the following manner:

[0035] The first fault characteristic parameters acquired in real time Input the digital twin model;

[0036] The digital twin model is driven to simulate fault states in a fault database and output a set of simulated fault parameters. ;in, To simulate fault types, To simulate the degree of matching;

[0037] Calculate the characteristic distance between the actual first fault characteristic parameter and each simulated fault characteristic parameter. ;

[0038] The simulated fault type with the smallest feature distance is identified. This is determined to be a fault diagnosis result. Preferably, the second fault characteristic parameters are generated based on the simulated fault type. The corresponding feature distance 𝐷, combined with the predicted remaining useful life RUL of the fault output by the digital twin model, together generate a comprehensive second fault feature parameter triplet, which is used to support operation and maintenance judgment decisions.

[0039] Preferably, the system further includes an online model update module, used for:

[0040] After repair, confirm whether the actual fault is consistent with the first fault diagnosis result and the second fault diagnosis result;

[0041] When the actual fault is inconsistent with the first and second fault diagnosis results and exceeds the preset threshold, the new fault data sample and its finally confirmed fault label are automatically added to the training set, triggering the incremental learning process of the first-level fault diagnosis model and updating the model parameters.

[0042] The beneficial effects of this invention are:

[0043] (1) This invention achieves adaptive data acquisition under different operating conditions by following preset variable speed sampling rules, avoiding the acquisition of a large amount of redundant data during stable operation, thus saving storage space and transmission bandwidth; at the same time, the transient high-frequency sampling mode ensures that the detailed features of the signal (such as the waveform and spectrum of the impact) can be captured with the highest accuracy during the period of potential faults or sudden anomalies, providing a crucial data foundation for the accurate diagnosis of early faults.

[0044] (2) Through principal component analysis (PCA) dimensionality reduction and fuzzy C-means clustering (FCM), the system can extract the main operating condition features from complex operating data in an unsupervised and automated manner. Furthermore, this refined processing greatly improves the adaptability and diagnostic accuracy of the fault diagnosis model under varying operating conditions, and solves the problem of the performance of traditional fixed thresholds or single models dropping sharply when operating conditions change.

[0045] (3) This invention uses a data processing module to quickly identify faults in risk models under different operating conditions. A key innovation is that a dedicated deep convolutional neural network (CNN) is trained for each operating condition, and wavelet time-frequency plots are used as input. CNN can automatically learn richer and deeper fault patterns from time-frequency plots than traditional manual features (such as peaks and kurtosis); it can also identify weak early fault features in advance, so that fault diagnosis has a basis in advance; by calculating the matching degree M between the probability distribution of the model output and the preset threshold, the fault diagnosis result is transformed from a simple "yes / no" judgment into a continuous and quantifiable risk indicator; so that maintenance personnel can not only know "what faults may occur", but also "how likely the fault is" and "how serious the fault is", thus providing more refined and forward-looking data support for predictive maintenance decisions.

[0046] (4) This invention integrates historical and multi-condition information into a health index through a data operation and maintenance module to achieve an intuitive assessment of the macroscopic status of the equipment; specifically, it ensures that multiple fault risk signals scattered at different times and under different conditions are integrated into a single, intuitive comprehensive indicator through standard health index calculation; the index introduces dynamic weights based on maintenance costs and frequency, making the assessment results more in line with the actual economic and safety requirements of equipment management; and managers can grasp the overall health degradation of the core components of the equipment in real time based on the trend of the health index.

[0047] (5) The present invention also inputs the data-driven primary diagnostic results (first fault feature parameters) into the digital twin model for simulation comparison, thereby integrating data-driven and model-driven approaches and improving the execution of diagnostic results; by calculating the feature distance D, the severity of the fault can be quantified more accurately, and even different stages of fault development can be distinguished; the digital twin model can predict the remaining useful life (RUL) based on the current state and generate comprehensive operation and maintenance decisions. Through the generated second fault feature parameter triplet (fault type, feature distance, remaining life), it provides richer information depth for operation and maintenance decisions; ensuring that operation and maintenance personnel not only know the fault location but also can judge the fault service life in a timely manner, thereby achieving risk operation and maintenance more stably and reducing operation and maintenance costs.

[0048] Of course, any product implementing this invention does not necessarily need to achieve all the advantages described above at the same time. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a block diagram of an intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1 As shown, the present invention is an intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment, comprising:

[0053] The equipment-layer sensor array is deployed in key parts of low-speed, heavy-load equipment; the equipment-layer sensor array includes at least two of the following: vibration acceleration sensors, acoustic emission sensors, temperature sensors, and speed sensors;

[0054] The data acquisition module communicates with the sensor array at the device layer and is used to acquire equipment operating status data of key parts in the sensor array according to preset sampling rules; the equipment operating status data includes at least two of the following: vibration signal, acoustic emission signal, temperature signal, and rotational speed signal;

[0055] The data processing module is used to identify the operating condition type of the preprocessed equipment operating status data, cluster and extract multiple operating conditions of the equipment operating status, perform machine learning on each operating condition, pre-train and obtain the first-level fault diagnosis model associated with the operating conditions, and input real-time data into the first-level fault diagnosis model for feature extraction, and determine the first fault feature parameters based on the fault matching degree of the corresponding operating conditions.

[0056] The data operation and maintenance analysis module is used to receive the first fault characteristic parameters, perform fusion analysis on the first fault characteristic parameters under various historical operating conditions to obtain the standard health index of the core components of the equipment; compare the actual monitored first fault characteristic parameters with the simulated fault characteristic parameters of the digital twin model under the fault state to obtain the fault diagnosis results, generate the second fault characteristic parameters, and make operation and maintenance judgment decisions based on the fault diagnosis results.

[0057] In the above technical solution, firstly, multi-sensor fusion and adaptive clustering technology are used to dynamically identify and classify the operating conditions of the equipment; then, a dedicated deep learning diagnostic model is trained for each operating condition to achieve accurate extraction of primary fault features; next, digital twin technology is used to simulate and verify the primary diagnostic results in a virtual space and conduct in-depth analysis to generate comprehensive parameters that integrate fault type, severity, and remaining life prediction; finally, through an online update module, the system can continuously optimize the diagnostic model using newly generated data to form a highly adaptive intelligent system. Example

[0058] First, the system architecture and equipment-level sensor deployment are established. An equipment-level sensor array, designed for low-speed, heavy-load, multi-condition equipment (such as a ball mill), is deployed at key locations within the ball mill. Vibration acceleration sensors are installed on the main bearing housing and pinion bearing housing, with measurement points set. Acoustic emission sensors (resonant frequency 150kHz) are installed in the middle of the cylinder and near the feed end bearing, with measurement points set, to monitor low-frequency, high-energy events such as cracks and liner wear. Temperature sensors (PT100) are installed at the main bearing lubrication outlets (2) and the motor bearings (2), with measurement points set. Speed ​​sensors (Hall effect) are installed on the drive shaft, with measurement points set, to accurately acquire speed and transient processes during start-up and shutdown. The above are the number of measurement points.

[0059] Then, the data acquisition module communicates with all sensors to collect and cache data in real time, enabling the acquisition of equipment operating status data. The acquisition adopts preset sampling rules: First, during the start / stop phase (speed from 0 to 90% of the steady-state rated speed or vice versa): the sampling frequency is initially set to 10kHz (vibration / acoustic emission) to capture transient processes, continuing until the speed stabilizes; Second, during the steady-state operation phase: the sampling frequency is reduced to 1 to 2kHz (vibration / acoustic emission); Third, safety threshold triggering: a vibration acceleration safety threshold is set; when the amplitude of any vibration channel instantaneously exceeds the vibration acceleration safety threshold, a transient high-frequency sampling mode is automatically triggered (vibration / acoustic emission sampling rate increases to 20kHz, lasting 2-3 seconds, to capture the details of abnormal impacts).

[0060] Next, the data processing module is deployed on a cloud server to receive preprocessed data from the edge gateway. The module first performs preprocessing on the raw signal, including denoising and normalization. Then, it constructs a comprehensive operating condition feature vector based on a feature extraction method using multi-sensor data fusion. Principal component analysis (PCA) is then used to reduce the dimensionality of this vector. Specifically:

[0061] Constructing a comprehensive working condition feature vector ;in, The dominant frequency of vibration, It is the root mean square value; It's temperature. It's the rotational speed; The root mean square value of the vibration signal. This is the root mean square value of the transmitted signal; For the bearing temperature, principal component analysis (PCA) is used to analyze the vector... Dimensionality reduction is performed, and the two principal components with a cumulative contribution rate greater than 85% are used as clustering inputs;

[0062] The optimal number of operating condition clusters for principal components is determined using the silhouette coefficient method. ;set up ,in The number 3 can represent light load low speed, heavy load high speed, and transitional working conditions, respectively.

[0063] The Fuzzy C-means Clustering (FCM) algorithm is used to calculate the operating status data samples for each device. For each working condition cluster membership degree ∈ And satisfy The operating condition type of each device's operating status data sample is determined by its maximum membership degree. The corresponding operating condition clusters are determined.

[0064] Training of the first-level fault diagnosis model:

[0065] A separate deep convolutional neural network (CNN) model is built for each operating condition cluster.

[0066] Model input: The original vibration signal is converted into a time-frequency diagram through continuous wavelet transform to preserve both time-domain and frequency-domain features.

[0067] Model Output: The output layer uses the Softmax function to output the probability distributions corresponding to the four states of the ball mill under this operating condition: "normal", "bearing inner ring failure", "bearing outer ring failure", and "gear tooth breakage". Historical data (including labels) under various operating conditions are used to train the corresponding CNN model until the model converges. Historical data under each operating condition and its corresponding fault labels are extracted, and a supervised learning algorithm is used to train the model, ultimately generating a cluster of operating conditions. The corresponding first-level fault diagnosis model.

[0068] Furthermore, real-time diagnosis and health assessment are performed. First, the primary fault characteristic parameters are determined. After wavelet transform, the real-time data is input into the CNN model of its corresponding operating condition to obtain the probability distribution of the real-time data. Preset fault probability threshold vector (No threshold required); the probability distribution of this real-time data With the preset fault probability threshold vector ,and The matching degree is calculated using the following formula: Substituting the data from this implementation, then ;when When the equipment is determined to have a fault tendency, the fault type that reaches the maximum value is "bearing inner ring fault", and this will cause... Real-time fault type that obtains the maximum value and its real-time matching degree Both are used as the first fault characteristic parameters The output thus generates the first fault characteristic parameters. .

[0069] By fusing and analyzing the primary fault characteristic parameters under various historical operating conditions, a standard health index for the core components of the equipment can be obtained. The calculation process is as follows:

[0070] Collect the set of first fault characteristic parameters under various operating conditions over a historical period;

[0071] Feature fusion is performed on the first set of fault characteristic parameters to obtain the standard health index. The weighted calculation formula is used to obtain the following:

[0072]

[0073] in, This represents the total number of fault types. Fault type Average matching degree over a historical period; Fault type The weighting coefficient is determined based on the historical frequency of the fault and the cost of fault repair.

[0074] Standard health index calculation:

[0075] Collect the first fault characteristic parameters of all operating conditions in the past 24 hours and calculate the average matching degree of various fault types. ,For example ; ; ;

[0076] Weights are assigned based on maintenance cost and frequency. ,For example , ; ;

[0077] Calculate the standard health index:

[0078]

[0079] Principal component analysis was used to reduce the dimensionality of the first fault characteristic parameters under various historical operating conditions. Then, the weighted average method was used to fuse the scores of each principal component. The weighting coefficients were determined according to the frequency and severity of the faults under each operating condition. Finally, a standard health index between 0 and 1 was obtained, where 1 represents complete health and 0 represents severe fault.

[0080] In addition, through digital twin verification and integrated decision-making, the second fault diagnosis result is presented in the following manner:

[0081] The first fault characteristic parameters acquired in real time Input the digital twin model;

[0082] Driving digital twin models in fault databases The system simulates fault conditions and outputs a set of simulated fault parameters. ;in, To simulate fault types, To simulate the matching degree, the digital twin model is driven to simulate states such as "mild bearing inner ring fault" and "moderate bearing inner ring fault" in the fault database, and the simulated parameters are output as the second fault characteristic parameters, such as... , ;

[0083] Calculate the characteristic distance between the actual first fault characteristic parameter and each simulated fault characteristic parameter. ; Calculate feature distance ;

[0084] The simulated fault type with the smallest feature distance is identified. This was determined to be a fault diagnosis result; minimum feature distance. The corresponding simulated fault type is "mild bearing inner ring fault"; this is the second fault diagnosis result, which has a higher confidence level.

[0085] Based on the simulated fault type The corresponding feature distance φ, combined with the predicted remaining useful life (RUL) of the fault output by the digital twin model, together generate a comprehensive triplet of second fault feature parameters. This parameter is used to support operational and maintenance decision-making. Specifically, to generate the second fault feature parameters, the digital twin model, based on the physical model and damage accumulation theory, predicts that the remaining useful life (RUL) of the bearing at this fault level is 480 hours. Finally, the triplet of second fault feature parameters is generated. ;

[0086] Therefore, maintenance personnel can make maintenance decisions directly based on this information, and achieve predictive maintenance by replacing the bearings during the most recent planned shutdown after the equipment has been running for 400 hours (with a safety margin), thereby avoiding unplanned downtime.

[0087] Finally, the system includes an online model update module for continuous operation. For example, after a repair, the fault is confirmed to be a "bearing ball fault," while the system's initial first and second diagnostic results are both "inner race fault." When this inconsistency accumulates to more than a preset threshold number of repairs (e.g., 3 times), the system will automatically add the vibration time-frequency graph of that event and the "ball fault" label to the training set, triggering an incremental learning process to fine-tune and update the CNN model for the corresponding operating condition. This will enable the system to correctly identify such new faults in the future, improving recognition accuracy.

[0088] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0089] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended documents. In some cases, the actions or steps described in this application may be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0090] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this application, they should all fall within the protection scope of the present invention.

Claims

1. A smart operation and maintenance system for low-speed, heavy-load, multi-condition equipment, characterized in that, include: A device-layer sensor array is deployed in key parts of low-speed, heavy-load equipment; the device-layer sensor array includes at least two of the following: vibration acceleration sensors, acoustic emission sensors, temperature sensors, and speed sensors; The data acquisition module is communicatively connected to the sensor array of the device layer and is used to acquire equipment operating status data of key parts in the sensor array according to preset sampling rules; the equipment operating status data includes at least two of the following: vibration signal, acoustic emission signal, temperature signal, and rotational speed signal; The data processing module is used to identify the operating condition type of the preprocessed equipment operating status data, cluster and extract multiple operating conditions of the equipment operating status, and obtain the first-level fault diagnosis model associated with the operating conditions by performing machine learning on each operating condition and pre-training. And input real-time data into the first-level fault diagnosis model for feature extraction, and determine the first fault feature parameters based on the fault matching degree of the corresponding working condition; The data operation and maintenance analysis module is used to receive the first fault characteristic parameters, perform fusion analysis on the first fault characteristic parameters under various historical operating conditions to obtain the standard health index of the core components of the equipment; compare the actual monitored first fault characteristic parameters with the simulated fault characteristic parameters of the digital twin model under the fault state to obtain the fault diagnosis results, generate the second fault characteristic parameters, and make operation and maintenance judgment decisions based on the fault diagnosis results. The method for determining the first fault characteristic parameter is as follows: Probability distribution of obtaining this real-time data With the preset fault probability threshold vector ,and The matching degree is calculated as follows: ； when At that time, it is determined that the equipment is prone to failure, and will cause The fault type that achieves the maximum value and its matching degree are both used as the first fault feature parameter. Output; where, This is a real-time fault type; For real-time matching accuracy; For real-time probability distribution; The real-time value of the preset fault probability threshold vector; The method involves fusing and analyzing the first fault characteristic parameters under various historical operating conditions to obtain the standard health index of the equipment's core components. The calculation process is as follows: Collect the set of first fault characteristic parameters under various operating conditions over a historical period; The first fault feature parameter set is subjected to feature fusion, and the standard health index is... The weighted calculation formula is used to obtain the following: ； in, This represents the total number of fault types. Fault type The average matching degree over a certain period of history; Fault type The weighting coefficient is determined based on the historical frequency of the fault and the cost of fault repair. The method involves pre-training a first-level fault diagnosis model through machine learning for each operating condition. The specific approach is as follows: For each operating condition cluster A dedicated deep convolutional neural network model is established and trained separately. The input of the model is the time-frequency diagram of the equipment vibration signal obtained after wavelet transform processing. The output layer of the model uses the Softmax function to output the probability distribution of the equipment in various fault states under this operating condition. .

2. The intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to claim 1, characterized in that, The preset sampling rules include: The sampling frequency is set according to the equipment operation cycle. High-frequency sampling is used during the equipment start-up and shutdown phases, and low-frequency sampling is used during the stable operation phase. When the data acquisition module detects that the signal amplitude exceeds the preset safety threshold, the transient high-frequency sampling mode is automatically triggered.

3. The intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to claim 1, characterized in that, The method for identifying the operating condition type of the preprocessed equipment operating status data is as follows: a feature extraction method based on multi-sensor data fusion is used to construct a comprehensive operating condition feature vector; and principal component analysis is used to reduce the dimensionality of the comprehensive operating condition feature vector, with the dimensionality-reduced principal components used as the input of the clustering algorithm.

4. The intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to claim 3, characterized in that, The various operating conditions of the cluster extraction device are as follows: The optimal number of operating condition clusters for the principal components is determined using the silhouette coefficient method. ; The fuzzy C-means clustering algorithm was used to calculate the operating status data samples for each device. For each working condition cluster membership degree ∈ And satisfy ; The operating condition type of each device's operating status data sample is determined by the operating condition cluster corresponding to its maximum membership degree.

5. The intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to claim 1, characterized in that, The second fault diagnosis result is presented in the following manner: The first fault characteristic parameters acquired in real time Input the digital twin model; Driving digital twin models in fault databases The system simulates fault conditions and outputs a set of simulated fault parameters. ;in, To simulate fault types, To simulate the degree of matching; Calculate the characteristic distance between the actual first fault characteristic parameter and each simulated fault characteristic parameter. ; The simulated fault type with the smallest feature distance is identified. This is confirmed as the fault diagnosis result.

6. The intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to claim 5, characterized in that, The generation of the second fault characteristic parameters is based on the simulated fault type. The corresponding feature distance φ, combined with the predicted remaining useful life (RUL) of the fault output by the digital twin model, together generate a comprehensive second fault feature parameter triplet, which is used to support the operation and maintenance judgment decision.

7. The intelligent operation and maintenance system for low-speed, heavy-load, multi-condition equipment according to claim 1, characterized in that, The system also includes an online model update module, used for: After repair, confirm whether the actual fault is consistent with the first fault diagnosis result and the second fault diagnosis result; When the actual fault is inconsistent with the first and second fault diagnosis results and exceeds the preset threshold, the new fault data sample and its finally confirmed fault label are automatically added to the training set, triggering the incremental learning process of the first-level fault diagnosis model and updating the model parameters.

Images