A method for monitoring a failure of a stacker-reclaimer and a system therefor

Abstract

This invention provides a fault monitoring method and system for stacker-reclaimers, relating to the technical field of stacker-reclaimer fault monitoring. The method includes acquiring first information, which includes operating parameter information and structural parameter information of the stacker-reclaimer equipment; performing data filtering, noise reduction, and time-series alignment on the first information to obtain preprocessed first information; inputting the preprocessed first information into a preset fault feature extraction model for processing to obtain a fault feature set; classifying the equipment's operating status and calculating anomaly confidence based on the fault feature set to obtain second information, which includes the equipment's operating status classification result and anomaly confidence value; establishing a fault prediction model based on the second information; and predicting equipment fault prediction results based on the fault prediction model. This invention can more effectively improve the operational stability and maintenance efficiency of stacker-reclaimers.

Description

Technical Field

[0001] This invention relates to the technical field of stacker-reclaimer fault monitoring, and more specifically, to a fault monitoring method and system for stacker-reclaimers. Background Technology

[0002] Stacker-reclaimers are key equipment in bulk material handling systems, and their operating status directly affects production efficiency and equipment lifespan. However, due to prolonged high-load operation, stacker-reclaimers are prone to mechanical failures, electrical faults, and transmission system malfunctions. Existing fault monitoring methods mainly rely on threshold settings and rule-based expert systems for monitoring. Faults are identified by collecting equipment operating parameters and determining whether preset thresholds are exceeded. However, this method has significant limitations: firstly, threshold setting is subjective and difficult to adapt to complex operating conditions, leading to numerous false alarms or missed alarms; secondly, rule-based expert systems rely on human experience to formulate rules, making it difficult to effectively handle high-dimensional, multimodal equipment operating data and accurately predict fault occurrences.

[0003] Therefore, there is an urgent need for a fault monitoring method and system for stacker-reclaimers to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a fault monitoring method and system for stacker-reclaimers to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] In a first aspect, the present invention provides a fault monitoring method for a stacker-reclaimer, comprising:

[0006] Obtain first information, which includes the operating parameter information and structural parameter information of the stacker-reclaimer equipment;

[0007] The first information is subjected to data filtering, noise reduction and time alignment to obtain the preprocessed first information;

[0008] The preprocessed first information is input into a preset fault feature extraction model for processing to obtain a fault feature set;

[0009] Based on the fault feature set, the equipment operating status is classified and the anomaly confidence level is calculated to obtain second information, which includes the equipment operating status classification result and the anomaly confidence value.

[0010] A fault prediction model is established based on the second information, and the equipment fault prediction result is obtained based on the fault prediction model.

[0011] Secondly, the present invention also provides a fault monitoring system for a stacker-reclaimer, comprising:

[0012] The acquisition unit is used to acquire first information, which includes operating parameter information and structural parameter information of the stacker-reclaimer equipment.

[0013] The first processing unit is used to perform data filtering, noise reduction and time alignment processing on the first information to obtain preprocessed first information.

[0014] The second processing unit is used to input the preprocessed first information into a preset fault feature extraction model for processing to obtain a fault feature set;

[0015] The calculation unit is used to classify the equipment operating status and calculate the anomaly confidence level based on the fault feature set to obtain second information, which includes the equipment operating status classification result and the anomaly confidence value.

[0016] The prediction unit is used to establish a fault prediction model based on the second information, and to predict the equipment fault prediction result based on the fault prediction model.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention employs techniques such as manifold regularization, differential entropy analysis, and dynamic time warping to efficiently preprocess operational data and extract high-quality fault features. Secondly, it utilizes spectral analysis, tensor embedding, and graph-structured attention networks to construct a deep feature extraction model, enhancing the representational ability of fault features. Furthermore, it combines conformal prediction residual networks and variational mode decomposition techniques to perform high-confidence classification of equipment operating states and calculates anomaly confidence using a sparse Bayesian classifier, improving the robustness of fault detection. In addition, this invention proposes a fault prediction method based on generalized state-space variational inference and combines particle swarm optimization and quantum genetic algorithms to optimize maintenance scheduling schemes, achieving precise equipment health management. Finally, it uses causal reasoning based on graph neural networks to analyze root causes and combines generative adversarial networks optimized by structural entropy to improve anomaly detection capabilities, while Bayesian hyperparameter optimization enhances the long-term adaptability of fault prediction. Moreover, this invention has significant advantages in fault monitoring accuracy, fault prediction foresight, and adaptive optimization of maintenance strategies, effectively improving the operational stability and maintenance efficiency of stacker-reclaimers.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the fault monitoring method for the stacker-reclaimer described in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the fault monitoring system of the stacker-reclaimer described in an embodiment of the present invention.

[0023] In the figure: 701, acquisition unit; 702, first processing unit; 703, second processing unit; 704, calculation unit; 705, prediction unit. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a fault monitoring method for a stacker-reclaimer.

[0028] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0029] Step S1: Obtain first information, which includes the operating parameter information and structural parameter information of the stacker-reclaimer equipment;

[0030] It is understood that the operational parameter information in this step typically includes motor current, voltage, speed, torque, temperature, vibration signals, and hydraulic system pressure. These parameters can reflect the equipment's working status and potential anomalies in real time. For example, abnormal vibration signals may indicate wear on bearings or gears, while fluctuations in current and torque may be related to overload or transmission system failure. Furthermore, structural parameter information mainly involves the mechanical design characteristics of the stacker-reclaimer, such as the dimensions of the main components, material properties, and the topology of the moving mechanisms. This invention employs distributed sensor networks or Industrial Internet of Things (IIoT) technology for data acquisition and processes it in real time through edge computing devices or cloud data platforms. In addition, to improve data reliability, it is necessary to combine sensor redundancy design and data verification mechanisms to reduce the impact of signal interference and sensor failures on the monitoring results. The technical effect of this step is to provide high-quality basic data, enabling subsequent feature extraction, status assessment, and fault prediction to be carried out based on accurate data information, thereby improving the monitoring accuracy and real-time performance of the entire system.

[0031] Step S2: Perform data filtering, noise reduction, and time-series alignment on the first information to obtain the preprocessed first information;

[0032] It is understandable that this step, through data filtering, noise reduction, and time-series alignment, can effectively improve data quality, making it more representative, stable, and time-consistent. It can also reduce noise interference, improve data reliability, remove redundant information, reduce computational complexity, ensure data time synchronization, and improve the accuracy of subsequent fault analysis and prediction models, providing high-quality input data for intelligent fault monitoring of the stacker-reclaimer. In this step, step S2 includes steps S21, S22, and S23.

[0033] Step S21: Evaluate the manifold structure of the first information by constructing a manifold regularization model, filter out the first information whose representativeness is greater than a preset first threshold, and obtain the filtered first information;

[0034] Understandably, the core idea of ​​this step, the manifold regularization model, is to utilize the manifold structure of the data for optimization and filtering. Since the operational data of a stacker-reclaimer is often in a high-dimensional space, while its true key information may be distributed on a low-dimensional manifold, manifold learning methods can be used to find the low-dimensional representation of the data, thereby retaining the most representative feature information. In this step, a locally linear embedding method is used to perform nonlinear dimensionality reduction on the first piece of information, extract the manifold structure of the data, and calculate the geometric importance of each data point on the manifold.

[0035] Secondly, representativeness assessment is performed using regularization methods. After obtaining a low-dimensional representation of the initial information, we define a manifold regularization objective function to calculate the importance score of the data points. The regularization objective function is shown below:

[0036]

[0037] Where L(Z, μ) represents the objective function, W ij Represents sample point X i and sample point X j The manifold weights between them, where λ represents the weight coefficient of the regularization term, B(X) i ) represents the regularization function, μ i Represents the Lagrange multipliers. Represents sample point X i The target value.

[0038] Finally, data is filtered based on a first threshold. After calculating the representative score for each data point, we set a preset first threshold and filter out data points with representative scores greater than this threshold, forming the filtered first information. It is understandable that filtering the first information can effectively improve data quality, providing more representative and reliable input data for subsequent noise reduction, feature extraction, and fault analysis.

[0039] Step S22: Determine the stationarity value of the first information after filtering by calculating its differential entropy, and smooth the first information with a stationarity value less than the preset second threshold by using Gaussian mixture filtering to obtain the first information after noise reduction.

[0040] It is understandable that the core objective of this step is to assess the stationarity of the data based on differential entropy and to smooth the non-stationary data through Gaussian mixture filtering in order to reduce noise interference and improve the usability of the data.

[0041] First, the differential entropy is calculated to assess the stationarity of the data. The differential entropy in this step is defined as follows:

[0042] H(X)=-∫p(x)log p(x)dx

[0043] Where H(X) represents the differential entropy of dataset X, p(x) represents the probability density function of random variable x, and log p(x) represents the information content corresponding to probability density p(x).

[0044] In this step, we calculate the differential entropy value within the sliding window of the first information after filtering, and set a stationarity criterion: if the differential entropy is large, it indicates that the data has high uncertainty and exhibits strong non-stationarity; conversely, if the differential entropy is small, the data is relatively stable.

[0045] Then, Gaussian kernels with different standard deviations are selected to accommodate signals with different frequency components. The convolution results of multiple Gaussian kernels are then fused according to weights to ensure that high-frequency noise is smoothed while preserving the main trend of the original signal. The calculation formula is as follows:

[0046]

[0047] Among them, X filtered This represents the filtered output signal or data, where X represents the dataset. The function represents a filtering function, * represents a convolution operation, and w i This represents the weight of each filtering result. This represents summing over n different filters.

[0048] This step assesses the stationarity of the data using differential entropy and smooths non-stationary data using Gaussian mixture filtering, ultimately obtaining the first information after denoising. This method improves the stationarity of the data while preserving the main trend of the signal during denoising, thus ensuring the reliability of subsequent modeling and analysis.

[0049] Step S23: Adjust the time difference of the first information after noise reduction based on the dynamic time warping algorithm, and perform nonlinear time scale transformation on the adjusted first information through the variational autoencoder to obtain the preprocessed first information.

[0050] Understandably, this step first calculates the local matching distance between time series. For two time series, a cumulative distance matrix is ​​constructed, where each element represents the cumulative minimum cost of the current match. Then, the optimal alignment path is found through a backtracking algorithm to minimize the difference between the time series. This allows the time scale of the first information after noise reduction to be adjusted to align with the baseline time series.

[0051] Then, the adjusted time series features are extracted using an LSTM neural network, mapping the high-dimensional time series to a low-dimensional latent variable space. Variational modeling is performed using the normal distribution assumption to ensure the continuity and controllability of the data. The time series is reconstructed from the latent space using another set of neural networks, and the time scale is nonlinearly transformed to generate a time series optimized by the time scale.

[0052] Finally, the loss function is jointly optimized using reconstruction error loss and KL divergence loss to obtain the preprocessed first information. This step combines dynamic time warping algorithm and variational autoencoder to adjust the time difference and perform nonlinear time scale transformation on the denoised first information, thereby obtaining the preprocessed first information, making the data more consistent in the time domain, and providing a more stable and accurate input for subsequent intelligent analysis.

[0053] Step S3: Input the preprocessed first information into a preset fault feature extraction model for processing to obtain a fault feature set;

[0054] It is understandable that this step employs deep learning or signal processing methods to improve the ability to identify complex fault modes. Through dimensionality reduction and automatic filtering, irrelevant information is reduced, computational efficiency is improved, and higher-quality data input is provided for subsequent analysis. In this step, step S3 includes steps S31, S32, and S33.

[0055] Step S31: Perform spectrum analysis on the preprocessed first information, wherein frequency features are extracted using non-uniform Fourier transform to obtain a preliminary spectrum feature set;

[0056] It is understandable that this step uses a time window segmentation method to slice the preprocessed first information to adapt it to the requirements of spectrum analysis and ensure that the data format meets the calculation requirements of non-uniform Fourier transform. This step calculates the spectrum of non-uniform data using the following mathematical expression:

[0057]

[0058] Where X(f) represents the spectrum of signal x(t) at frequency f, x(t) m ) indicates that the signal is at time point t m The value on, The complex exponent is a key term in the Fourier transform used to convert a time-domain signal to a frequency-domain signal; j is the imaginary unit, 2 π ft m Representing frequency f and time point t m The exponential term in the relationship between M and m represents the total number of sampling points of the signal and the time index.

[0059] This step calculates the spectral power spectral density, extracts the main frequency components, and then selects high-energy peak frequencies to avoid low-energy noise interference, obtaining the main frequency components and their corresponding energy values ​​to form a feature vector. By extracting high-energy peak frequencies, this step can distinguish between normal operating conditions and fault conditions, improving diagnostic capabilities. Non-uniform Fourier transform is suitable for irregular data sampling scenarios, avoiding the additional calculations of data interpolation or resampling, thus improving efficiency.

[0060] Step S32: Input the preliminary spectral feature set into the sparse tensor manifold embedding model for processing. The preliminary spectral feature set is transformed into a high-order tensor and the feature dimensionality is reduced by using the manifold learning method to obtain the dimensionality-reduced fault feature set.

[0061] Understandably, this step converts the initial spectral feature set into a tensor representation, then normalizes it to ensure data range consistency and improve model stability. Next, it uses Laplacian eigenmaps to construct a data adjacency graph and calculate the local geometric structure of the data. Through local preserving projection, it finds the projection direction in the reduced-dimensional space that best retains the local structure of the original data. Finally, it employs manifold regularization to ensure that the features retain their class discriminative ability after dimensionality reduction, resulting in the dimensionality-reduced fault feature set. This step reduces redundant features and improves the computational speed of subsequent fault detection through tensor modeling and dimensionality reduction. The manifold embedding method preserves the nonlinear structure of high-dimensional data, making the dimensionality-reduced fault feature set more discriminative.

[0062] Step S33: Input the dimensionality-reduced fault feature set into a graph structure attention network to construct a graph structure model, and use the attention mechanism to extract global and local features to obtain the optimized fault feature set.

[0063] Understandably, this step uses the dimensionality-reduced fault feature set as nodes in a graph, with each node representing a feature. The similarity between features is calculated to establish adjacency relationships. Cosine similarity is used to generate an adjacency matrix. Then, the feature values ​​and the adjacency matrix are combined to form a graph signal. For each feature node, the attention weight is calculated using the following formula:

[0064]

[0065] Where, α ij The attention coefficients or weights between node i and node j are represented by ∑ k∈N(i) This represents summing the neighboring nodes N(i) of node i. This means performing an inner product operation between the parameter vector a and the concatenated feature vectors of nodes i and k. LeakyReLU is an activation function that has a very small slope for negative inputs and outputs itself for positive inputs.

[0066] Then, global and local features are weighted and aggregated to calculate new feature representations. This step can then iteratively compute an optimized fault feature set using a multi-layer graph-structured attention network. This step, through graph-structured modeling, preserves the topological information between fault features, making the features more expressive. Compared to traditional uniform weighting methods, the attention mechanism can adaptively adjust the weights of different features, reducing interference from unimportant features.

[0067] Step S4: Based on the fault feature set, classify the equipment operating status and calculate the anomaly confidence level to obtain the second information, which includes the equipment operating status classification result and the anomaly confidence level value;

[0068] It is understandable that this step combines optimized fault characteristics and a deep classification model to achieve accurate operational status classification and improve the reliability of equipment health monitoring. By calculating anomaly confidence, a health score for the equipment's operational status is provided to facilitate maintenance decisions. In this step, step S4 includes steps S41, S42, and S43.

[0069] Step S41: Process the fault feature set based on the conformal prediction residual network, wherein the equipment status is classified by the deep residual network and the confidence interval of each category is calculated by the conformal prediction theory to obtain the preliminary classification results and their confidence intervals;

[0070] It is understood that this step uses the fault feature set as input to the conformal prediction residual network, employing a preset number of residual blocks to ensure efficient gradient propagation, prevent gradient vanishing, and improve training stability. Then, local pattern information is extracted through convolutional layers to identify key features; next, a fully connected layer performs final classification, outputting the predicted probabilities of multiple device state categories. Finally, an activation function (Softmax) layer outputs the probability distribution of each category, yielding preliminary classification results. These are shown below:

[0071]

[0072] Wherein, P(B=b) c |A) indicates that, given input A, output B takes category b. c The conditional probability, This represents summing the index values ​​for all categories c. This represents the index value corresponding to a specific category b.

[0073] Next, this step calculates the confidence level of each category by using the probability distribution of each category, and constructs confidence intervals based on historical data to obtain the confidence intervals corresponding to different categories.

[0074] Step S42: Perform multi-level mode decomposition on the preliminary classification results and their confidence intervals based on the variational mode decomposition algorithm to obtain the decomposed fault mode features;

[0075] Understandably, this step involves constructing a variational mode decomposition model. This model uses a variational optimization framework to decompose the input signal (preliminary classification results and their confidence intervals) into several orthogonal mode components, each representing a specific fault mode. The variational mode decomposition model is then obtained by constructing the objective function shown below:

[0076]

[0077] Among them, h e(x) represents a signal or function on the new variable, g e Represents the function h e The frequency parameter corresponding to (x), This represents the operation on the partial derivative with respect to the variable x, where j represents the imaginary unit. Indicates signal h e The frequency domain representation of (x) Let f(x) represent the objective function f(x) in the frequency domain, α represent the regularization parameter, E represent the total number of signals or functions, and z represent the new variable used in the frequency domain.

[0078] Then, by setting the target number of modes, the number of decomposition layers is adaptively determined based on the spectral characteristics of the classification results. The alternating direction multiplier method is used for optimization iteration to obtain each modal component. Then, the spectral energy ratio of each mode is calculated to evaluate the importance of each mode. The mean, variance, kurtosis and skewness of each mode are calculated to analyze the distribution characteristics of different modes. The preliminary classification results and their confidence intervals are decomposed to obtain the decomposed fault mode features.

[0079] Step S43: Probabilistically model and calculate the probability confidence of the classification strategy of the decomposed fault mode features using a sparse Bayesian classifier to obtain the equipment operating status classification results and anomaly confidence.

[0080] Understandably, this step utilizes the sparse Bayesian method for probabilistic classification. Specifically, it uses Bayes' theorem to calculate the posterior probability of a device's state and performs sparse Bayesian inference to obtain the probabilistic classification result. Furthermore, it calculates the anomaly confidence level based on the probability distribution of each category. The sparse Bayesian method automatically filters key features, avoids feature redundancy, and improves the stability of the classification.

[0081] Step S5: Establish a fault prediction model based on the second information, and predict the equipment fault prediction result based on the fault prediction model.

[0082] It is understood that this step employs a probabilistic prediction method, providing not only the prediction result but also a confidence interval to quantify the uncertainty of the prediction and improve its reliability. The equipment failure prediction result includes the equipment's possible future operating states, the probability of failure at future time steps, and their confidence intervals. In this step, step S5 includes steps S51, S52, and S53.

[0083] Step S51: Establish a fault trend model based on the second information and the generalized state space variational inference algorithm, and optimize the state transition parameters of the fault trend model using the variational inference method to obtain the optimized fault trend model.

[0084] Understandably, this step represents the state of the second information evolving over time using a state-space model. Since the probability distribution of state transitions may be a complex nonlinear distribution, traditional maximum likelihood estimation is difficult to calculate. Therefore, variational inference is used to optimize the state transition parameters. Assuming the true posterior distribution of the system is complex and difficult to determine, a variational distribution is used for approximate estimation. Then, by optimizing and maximizing the variational lower bound, the optimal state transition parameters are found, enabling the model to more accurately describe the equipment's failure trend. Finally, gradient descent in the stochastic variational inference method is used to optimize the state transition parameters, resulting in the optimized failure trend model.

[0085] Step S52: Perform deep regression processing on the optimized fault trend model to obtain equipment health status prediction information under multiple preset time scales;

[0086] Understandably, this step uses a Long Short-Term Memory (LSTM) network for time series prediction. The LSTM output layer employs a multi-timescale regression head to predict short-term health status (1 hour, 6 hours, 12 hours), medium-term health status (1 day, 3 days, 7 days), and long-term health status (15 days, 30 days). Equipment health status often changes slowly over time; multi-scale prediction allows for advance planning of maintenance schedules, preventing sudden failures. For example, short-term prediction can provide early warnings of impending failures, medium-term prediction can help plan maintenance times, and long-term prediction can optimize equipment replacement cycles.

[0087] Step S53: Based on the partial differential equation dynamic fault model, process all equipment health status prediction information, and combine it with the finite difference method to calculate the remaining lifespan and future fault trend information of the equipment.

[0088] Understandably, this step utilizes a partial differential equation dynamic fault model to model equipment health status prediction information across multiple time scales, describing the spatiotemporal evolution of equipment health status. Subsequently, the finite difference method is used to numerically solve the partial differential equation dynamic fault model, calculating the changes in equipment health status over time through time stepping, and determining the time point when the health status reaches the fault threshold, thereby predicting the remaining lifespan of the equipment. Simultaneously, this step calculates the health status at different future time scales, obtaining the rate of decline of equipment health status and future fault trend information, further assessing the operational risks and maintenance needs of the equipment. Through the construction of the partial differential equation dynamic fault model and the numerical solution of the finite difference method, this step can accurately predict the remaining lifespan of the equipment. Compared to traditional statistical regression methods, this method can more accurately characterize the dynamic evolution process of health status. Step S5 is followed by steps S61, S62, and S63.

[0089] Step S61: Process the equipment failure prediction results and historical maintenance strategies through multi-agent reinforcement learning, and adjust the maintenance strategies of the agents through a reward mechanism to obtain a preliminary maintenance scheduling plan.

[0090] Understandably, this step first uses equipment failure prediction results and historical maintenance strategies as input to construct a maintenance optimization framework based on multi-agent reinforcement learning. Each agent corresponds to a different maintenance task or equipment subsystem and continuously optimizes its decisions through reinforcement learning during environmental interactions. A deep Q-network is used to enable each agent to iteratively update its maintenance strategy in a dynamic environment. By minimizing maintenance costs, reducing equipment downtime, or optimizing overall operational efficiency, the agents can continuously learn optimal maintenance decisions during training, ultimately forming a preliminary maintenance scheduling scheme. By introducing multi-agent reinforcement learning, this step can achieve adaptive optimization of maintenance strategies in complex maintenance scenarios, making maintenance decisions more flexible and adaptable. Compared to traditional rule-driven or fixed-model-based methods, reinforcement learning can fully utilize historical data and predictive information to achieve dynamic optimization.

[0091] Step S62: Perform a global search on the preliminary maintenance scheduling scheme based on the particle swarm optimization algorithm, and then perform a global optimization on the preliminary maintenance scheduling scheme in combination with the quantum genetic algorithm to obtain the optimized maintenance scheduling scheme;

[0092] Understandably, this step first utilizes the Particle Swarm Optimization (PSO) algorithm to perform a global search on the initial maintenance scheduling scheme, optimizing for local optima that the agent might get trapped in during reinforcement learning. The PSO algorithm simulates the collaborative search behavior of particles in the search space, adjusting the position and velocity of each particle to gradually converge to the optimal maintenance strategy. Building on this, a quantum genetic algorithm is further introduced. Through mechanisms such as quantum state encoding and quantum rotation gates, the scheduling scheme optimized by the PSO algorithm is globally optimized. The quantum genetic algorithm improves search efficiency through the characteristics of quantum superposition states and enhances the diversity of maintenance strategies using quantum mutation mechanisms, thereby avoiding getting trapped in local optima and obtaining a globally optimal maintenance scheduling scheme.

[0093] Step S63: Based on extension theory and the optimized maintenance scheduling scheme, construct a set of maintenance strategies, and dynamically adjust the maintenance strategies in combination with the equipment operating environment and operating conditions to obtain an adaptive maintenance strategy.

[0094] Understandably, this step employs extension theory to further refine the optimized maintenance scheduling scheme, constructing a set of maintenance strategies applicable to different operating conditions. First, based on the fundamental principles of extension theory, the maintenance scheduling scheme is mapped to an extension set, and multiple maintenance strategy categories are defined, such as preventative maintenance, predictive maintenance, and corrective maintenance. Then, the fitness of different maintenance strategies is calculated using correlation functions, and dynamic adjustments are made based on the equipment operating environment, operational conditions, and historical maintenance data. Specifically, the system dynamically optimizes the applicability of maintenance strategies by real-time monitoring of equipment status parameters (such as vibration, temperature, and load) and combining them with external operational conditions (such as production demands and maintenance resource availability), enabling it to adaptively select the optimal maintenance strategy. Ultimately, a flexible and adaptive maintenance strategy system is constructed to cope with the continuous changes in equipment operating conditions. This step, following step S63, also includes steps S64, S65, and S66.

[0095] Step S64: Based on equipment maintenance decisions and operating parameter information, construct a causal reasoning framework based on graph neural networks, analyze the causal relationships of different fault modes through the topological relationships between nodes, optimize the monitoring strategies of key sensors, and obtain the root cause analysis results of fault modes.

[0096] This step understandably treats equipment operating status, maintenance decisions, and sensor data as nodes in a graph. Through equipment physical structure, expert knowledge, and historical data analysis, an initial topology of causal relationships is established, defining which parameters might affect which failure modes. A graph attention network is used to calculate the influence weights between different sensor nodes, extracting key influence paths. A structural causal model is then combined with do-of-do (DO) operations to evaluate the causal impact of different nodes on failure modes. Finally, based on the learned causal relationships, the optimal explanatory paths for different failure modes are inferred, identifying the root causes of failures and providing accurate decision-making support for subsequent maintenance.

[0097] Step S65: Generate an adversarial network model based on the root cause analysis results of the fault modes, and optimize the adversarial network by adding a structural entropy regularization term to obtain an abnormal mode detection model.

[0098] Understandably, this step improves the model's anomaly detection capability in complex operating environments by generating features for failure modes through adversarial networks and utilizing structural entropy constraints to optimize data distribution. Specifically, by adding a structural entropy regularization term to the adversarial network model's loss function, the model not only focuses on learning the adversarial distribution of the data during optimization but also constrains the complexity of anomalous pattern data in high-dimensional space, preventing the model from getting trapped in local optima. The introduction of the structural entropy regularization term optimizes the learning process of the adversarial network, making the feature representation of anomalous patterns clearer, avoiding the model getting trapped in local optima, and thus improving the accuracy of anomalous pattern detection.

[0099] Step S66: Adaptively tune the hyperparameters of the anomaly pattern detection model based on the Bayesian optimization framework to obtain the anomaly pattern detection model.

[0100] Understandably, this step utilizes a Bayesian optimization framework to model the hyperparameter space through Gaussian process regression, selects the optimal hyperparameters for training based on the expectation enhancement criterion, and continuously updates the Gaussian process model, iteratively optimizing the search strategy. Finally, after reaching the set number of optimization rounds or when the detection performance converges, the optimized anomaly pattern detection model is output to improve the model's anomaly recognition ability and generalization performance. This effectively enhances the hyperparameter optimization efficiency of the anomaly pattern detection model, ensuring that the model has higher accuracy, robustness, and generalization ability in equipment anomaly monitoring and fault prediction tasks.

[0101] Example 2:

[0102] like Figure 2 As shown, this embodiment provides a fault monitoring system for a stacker-reclaimer. See [link to documentation]. Figure 2 The system includes an acquisition unit 701, a first processing unit 702, a second processing unit 703, a calculation unit 704, and a prediction unit 705.

[0103] The acquisition unit 701 is used to acquire first information, which includes the operating parameter information and structural parameter information of the stacker-reclaimer equipment;

[0104] The first processing unit 702 is used to perform data filtering, noise reduction and time sequence alignment processing on the first information to obtain preprocessed first information.

[0105] The second processing unit 703 is used to input the preprocessed first information into a preset fault feature extraction model for processing to obtain a fault feature set;

[0106] The calculation unit 704 is used to classify the equipment operating status and calculate the anomaly confidence level based on the fault feature set to obtain second information, the second information including the equipment operating status classification result and the anomaly confidence value;

[0107] The prediction unit 705 is used to establish a fault prediction model based on the second information and to predict the equipment fault prediction result based on the fault prediction model.

[0108] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0109] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0110] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fault monitoring method for a stacker-reclaimer, characterized in that, include: Obtain first information, which includes the operating parameters and structural parameters of the stacker-reclaimer equipment; The first information is subjected to data filtering, noise reduction and time alignment to obtain the preprocessed first information; The preprocessed first information is input into a preset fault feature extraction model for processing to obtain a fault feature set; Based on the fault feature set, the equipment operating status is classified and the anomaly confidence level is calculated to obtain second information, which includes the equipment operating status classification result and the anomaly confidence value. A fault prediction model is established based on the second information, and the equipment fault prediction result is obtained based on the fault prediction model. The process of classifying equipment operating status and calculating anomaly confidence based on the fault feature set includes: The fault feature set is processed based on a conformal prediction residual network, wherein the equipment status is classified by a deep residual network and the confidence interval of each category is calculated by conformal prediction theory to obtain preliminary classification results and their confidence intervals. Based on the variational mode decomposition algorithm, multi-level mode decomposition is performed on the preliminary classification results and their confidence intervals to obtain the decomposed fault mode features; By using a sparse Bayesian classifier to probabilistically model the classification strategy of the decomposed fault mode features and calculate the probability confidence, the classification results of equipment operating status and anomaly confidence are obtained. The process of establishing a fault prediction model based on the second information and predicting equipment fault prediction results based on the fault prediction model includes: A fault trend model is established based on the second information and the generalized state space variational inference algorithm. The state transition parameters of the fault trend model are then optimized using the variational inference method to obtain the optimized fault trend model. The optimized fault trend model is processed by deep regression to obtain equipment health status prediction information at multiple preset time scales. The partial differential equation dynamic fault model processes all equipment health status prediction information and combines it with the finite difference method to calculate the remaining lifespan and future fault trend information of the equipment.

2. The fault monitoring method for a stacker-reclaimer according to claim 1, characterized in that... The first information is then subjected to data filtering, noise reduction, and time-series alignment processing, including: The manifold structure of the first information is evaluated by constructing a manifold regularization model, and the first information with representativeness greater than a preset first threshold is filtered to obtain the filtered first information. The stationarity value of the first information after filtering is determined by calculating its differential entropy, and the first information with a stationarity value less than the preset second threshold is smoothed by Gaussian mixture filtering to obtain the first information after noise reduction. The time difference of the first information after noise reduction is adjusted based on the dynamic time warping algorithm, and the adjusted first information is transformed by nonlinear time scale through the variational autoencoder to obtain the preprocessed first information.

3. The fault monitoring method for a stacker-reclaimer according to claim 1, characterized in that... The preprocessed first information is input into a preset fault feature extraction model for processing, including: The preprocessed first information is subjected to spectrum analysis, in which frequency features are extracted using non-uniform Fourier transform to obtain a preliminary spectrum feature set; The preliminary spectral feature set is input into a sparse tensor manifold embedding model for processing. The preliminary spectral feature set is transformed into a high-order tensor and the feature dimensionality is reduced using a manifold learning method to obtain a dimensionality-reduced fault feature set. The dimensionality-reduced fault feature set is input into a graph structure attention network to construct a graph structure model, and an attention mechanism is used to extract global and local features to obtain an optimized fault feature set.

4. A fault monitoring system for a stacker-reclaimer, characterized in that, include: The acquisition unit is used to acquire first information, which includes operating parameter information and structural parameter information of the stacker-reclaimer equipment. The first processing unit is used to perform data filtering, noise reduction and time alignment processing on the first information to obtain preprocessed first information. The second processing unit is used to input the preprocessed first information into a preset fault feature extraction model for processing to obtain a fault feature set; The calculation unit is used to classify the equipment operating status and calculate the anomaly confidence level based on the fault feature set to obtain second information, which includes the equipment operating status classification result and the anomaly confidence value. The prediction unit is used to establish a fault prediction model based on the second information, and to predict the equipment fault prediction result based on the fault prediction model. The computing unit includes: The second calculation subunit is used to process the fault feature set based on the conformal prediction residual network, wherein the equipment status is classified by the deep residual network and the confidence interval of each category is calculated by the conformal prediction theory to obtain the preliminary classification results and their confidence intervals. The sixth processing subunit is used to perform multi-level mode decomposition on the preliminary classification results and their confidence intervals based on the variational mode decomposition algorithm to obtain the decomposed fault mode features. The third computational subunit is used to perform probabilistic modeling and calculate probability confidence of the classification strategy of the decomposed fault mode features through a sparse Bayesian classifier, so as to obtain the equipment operating status classification results and anomaly confidence. The prediction unit includes: The first prediction subunit is used to optimize the state transition parameters of the fault trend model using variational inference methods to obtain the optimized fault trend model. The second prediction subunit is used to process the optimized fault trend model through a deep regression process to obtain equipment health status prediction information under multiple preset time scales. The third prediction subunit is used to process all equipment health status prediction information based on the partial differential equation dynamic fault model, and calculate the remaining life and future fault trend information of the equipment by combining the finite difference method.

5. The fault monitoring system for the stacker-reclaimer according to claim 4, characterized in that, The first processing unit includes: The first processing subunit is used to evaluate the manifold structure of the first information by constructing a manifold regularization model, filter the first information whose representativeness is greater than a preset first threshold, and obtain the filtered first information. The first calculation subunit is used to determine the corresponding stationarity value by calculating the differential entropy of the filtered first information, and to smooth the first information with a stationarity value less than a preset second threshold by using Gaussian mixture filtering to obtain the first information after noise reduction. The second processing subunit is used to adjust the time difference of the first information after noise reduction based on the dynamic time warping algorithm, and to perform nonlinear time scale transformation on the adjusted first information through a variational autoencoder to obtain the preprocessed first information.

6. The fault monitoring system for a stacker-reclaimer according to claim 4, characterized in that, The second processing unit includes: The third processing subunit is used to perform spectrum analysis on the preprocessed first information, wherein frequency features are extracted using non-uniform Fourier transform to obtain a preliminary spectrum feature set. The fourth processing subunit is used to input the preliminary spectral feature set into the sparse tensor manifold embedding model for processing. The preliminary spectral feature set is transformed into a high-order tensor and the feature dimensionality is reduced by using the manifold learning method to obtain the dimensionality-reduced fault feature set. The fifth processing subunit is used to input the dimensionality-reduced fault feature set into the graph structure attention network to construct the graph structure model, and to extract global and local features using the attention mechanism to obtain the optimized fault feature set.

Images