Bucket wheel machine safety monitoring method and system

Abstract

The invention provides a bucket wheel machine safety monitoring method and system, and relates to the technical field of bucket wheel machine safety monitoring, and the method comprises the steps: obtaining a historical data set; performing feature extraction processing according to each period in the historical data set to obtain a high-dimensional feature vector set; performing graph construction processing according to the high-dimensional feature vector set to obtain a feature association graph set; performing embedding processing on the feature association graph set based on a preset graph neural network to generate a fusion feature vector set; modeling processing is carried out according to the fusion feature vector set, and an anomaly detection reference model is constructed; and inputting the real-time data into the anomaly detection reference model for comparison to obtain an anomaly detection result, the anomaly detection result including whether an anomaly exists or not and an anomaly classification result. Based on vibration, rotating speed, electrical signals and environmental parameters, the operation state of the bucket wheel machine is comprehensively described by combining time domain, frequency domain and multi-mode interaction characteristics, and the limitation of traditional single signal monitoring is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring of bucket wheel stacker reclaimers, and more particularly, to a safety monitoring method and system for bucket wheel stacker reclaimers. Background Art

[0002] With the development of industrial automation and intelligence, bucket wheel stacker reclaimers, as bulk material handling and conveying equipment, have been widely used in fields such as mines, power plants, and ports. During the operation of bucket wheel stacker reclaimers, their key components (such as main bearings, drive motors, etc.) are subjected to complex mechanical loads and environmental stresses. Real-time monitoring of vibration, rotational speed, electrical signals, and environmental parameters (such as dust concentration, humidity, temperature) is particularly important for the safe and stable operation of the equipment. Currently, traditional monitoring methods for bucket wheel stacker reclaimers mainly rely on limit alarm methods based on single-signal characteristics. For example, alarm thresholds are set by the amplitude change of vibration signals or simple monitoring is carried out on the range of rotational speed fluctuations. However, this method has the following problems: First, the monitoring of single signals cannot comprehensively reflect the interaction between components and systematic failures of bucket wheel stacker reclaimers under complex working conditions (such as frequent startup, stop, and commutation). Second, the fixed-threshold alarm mode is difficult to adapt to the dynamic changes of various operating conditions, and false alarms or missed alarms are likely to occur. Finally, due to the failure to fully consider the comprehensive impact of environmental parameters on equipment performance, existing methods have insufficient ability to identify abnormal phenomena caused by external environmental changes (such as high-dust and high-humidity conditions). Although some technologies have tried to introduce simple trend analysis based on historical data statistics or methods of multi-signal superposition in recent years, these methods still cannot achieve accurate diagnosis and prediction of the health status of equipment under complex operating environments.

[0003] Based on the above disadvantages of the existing technology, there is an urgent need for a safety monitoring method and system for bucket wheel stacker reclaimers. Summary of the Invention

[0004] The purpose of the present invention is to provide a safety monitoring method and system for bucket wheel stacker reclaimers to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] In a first aspect, the present application provides a safety monitoring method for bucket wheel stacker reclaimers, including:

[0006] Obtaining a historical data set, where the historical data set includes vibration signals, rotational speed signals, electrical signals, and corresponding environmental parameters collected during each historical operation cycle of the bucket wheel stacker reclaimer;

[0007] Performing feature extraction processing according to each cycle in the historical data set, and obtaining a high-dimensional feature vector set through analysis and quantization processing of the time domain, frequency domain, and multi-modal interaction features of various signals;

[0008] Perform graph construction processing based on the high-dimensional feature vector set, and obtain a set of feature association graphs by analyzing the physical correlation between different features and the actual influence degree of the operating environment;

[0009] Perform embedding processing on the set of feature association graphs based on a preset graph neural network, and generate a set of fused feature vectors through graph convolution operations and feature aggregation processing;

[0010] Perform modeling processing based on the set of fused feature vectors, and construct an anomaly detection benchmark model by performing clustering analysis on the fused feature vectors of different operating conditions. The anomaly detection benchmark model includes the feature center and standard deviation range of each operating condition;

[0011] Obtain the real-time data of the bucket wheel stacker-reclaimer, and input the real-time data into the anomaly detection benchmark model for comparison. By detecting the deviation degree of the real-time features, obtain the anomaly detection result, where the anomaly detection result includes whether there is an anomaly and the anomaly classification result.

[0012] In a second aspect, the present application also provides a safety monitoring system for a bucket wheel stacker-reclaimer, including:

[0013] An acquisition module that acquires a historical data set, where the historical data set includes vibration signals, rotation speed signals, electrical signals, and corresponding environmental parameters collected during each historical operation cycle of the bucket wheel stacker-reclaimer;

[0014] An extraction module for performing feature extraction processing according to each cycle in the historical data set, and obtaining a set of high-dimensional feature vectors by analyzing and quantifying the time domain, frequency domain, and multi-modal interaction features of various signals;

[0015] A construction module for performing graph construction processing based on the set of high-dimensional feature vectors, and obtaining a set of feature association graphs by analyzing the physical correlation between different features and the actual influence degree of the operating environment;

[0016] A fusion module that performs embedding processing on the set of feature association graphs based on a preset graph neural network, and generates a set of fused feature vectors through graph convolution operations and feature aggregation processing;

[0017] A modeling module for performing modeling processing based on the set of fused feature vectors, and constructing an anomaly detection benchmark model by performing clustering analysis on the fused feature vectors of different operating conditions. The anomaly detection benchmark model includes the feature center and standard deviation range of each operating condition;

[0018] A detection module, configured to obtain real-time data of a bucket wheel stacker-reclaimer, and input the real-time data into the anomaly detection benchmark model for comparison. By detecting the deviation degree of real-time features, an anomaly detection result is obtained, where the anomaly detection result includes whether there is an anomaly and the anomaly classification result.

[0019] The beneficial effects of the present invention are as follows:

[0020] Based on vibration, rotational speed, electrical signals, and environmental parameters, and combined with time domain, frequency domain, and multi-modal interaction features, the present invention comprehensively describes the operating state of the bucket wheel stacker-reclaimer, avoiding the limitations of traditional single-signal monitoring; through cluster analysis, the feature vectors under different working conditions are classified and modeled, and an anomaly detection benchmark model suitable for various working conditions such as startup, stable operation, and commutation is established, significantly improving the adaptability to dynamic working conditions and effectively reducing the phenomena of false alarms and missed alarms. Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use 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 thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a schematic flow chart of a safety monitoring method for a bucket wheel stacker-reclaimer described in an embodiment of the present invention;

[0023] Figure 2 It is a schematic structural diagram of a safety monitoring system for a bucket wheel stacker-reclaimer described in an embodiment of the present invention;

[0024] Figure 3 It is a schematic structural diagram of a safety monitoring device for a bucket wheel stacker-reclaimer described in an embodiment of the present invention.

[0025] Reference numerals in the figures: 800, a safety monitoring device for a bucket wheel stacker-reclaimer; 801, a processor; 802, a memory; 803, a multimedia component; 804, an I / O interface; 805, a communication component; 901, an acquisition module; 902, an extraction module; 903, a construction module; 904, a fusion module; 905, a modeling module; 906, a detection module. Detailed Embodiments

[0026] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0027] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0028] Embodiment 1:

[0029] This embodiment provides a safety monitoring method for a bucket wheel stacker-reclaimer.

[0030] See Figure 1 , which shows that this method includes steps S100 to S600.

[0031] Step S100: Obtain a historical data set, where the historical data set includes vibration signals, rotational speed signals, electrical signals, and corresponding environmental parameters collected during each historical operation cycle of the bucket wheel stacker-reclaimer;

[0032] It is understandable that vibration signals are collected by high-frequency acceleration sensors, usually installed at key components such as main bearings and speed reducers, to monitor the mechanical health status of the equipment and can reflect early problems such as main bearing wear and gear meshing faults; speed signals are obtained through non-contact photoelectric or magnetic induction speed sensors, directly reflecting the dynamic output state and operating conditions of the bucket wheel stacker-reclaimer, such as dynamic changes during startup, stop, or commutation; electrical signals include current, voltage, and harmonic distortion, which are collected through current transformers, voltage sensors, and power analyzers to judge the load characteristics and stability of the motor and drive system. For example, current fluctuations indicate motor overload or drive faults; environmental parameters such as dust concentration, humidity, and temperature are collected through laser dust sensors, temperature and humidity sensors, etc., reflecting the external conditions of equipment operation. High dust and high humidity may cause problems such as main bearing wear and electrical insulation failure. Through the comprehensive acquisition of multi-modal data, the historical dataset not only covers the operating characteristics of the equipment in the mechanical, electrical, and environmental dimensions but also provides complete data support for subsequent feature extraction and anomaly detection models, thus laying the foundation for the comprehensive analysis of equipment status and accurate fault diagnosis.

[0033] Step S200: Perform feature extraction processing according to each cycle in the historical dataset. By analyzing and quantifying the time-domain, frequency-domain, and multi-modal interaction features of various signals, a set of high-dimensional feature vectors is obtained.

[0034] It should be understood that time-domain feature extraction mainly quantifies the characteristics of signals such as amplitude, fluctuation, and change trend. For example, for vibration signals, the mean value, peak factor, kurtosis, etc. are extracted to reflect the overall change trend of main bearing vibration; for speed signals, the fluctuation amplitude and the limit speed during startup and stop are extracted to capture the dynamic changes of the power system. Frequency-domain features are extracted from signals through the fast Fourier transform (FFT) to obtain the main frequency, high-frequency energy ratio, etc., which are used to identify the vibration characteristics of the bucket wheel stacker-reclaimer under different working conditions and the harmonic characteristics of the motor drive. For example, harmonic distortion can reveal electrical system faults. The extraction of multi-modal interaction features combines the physical correlation between different signals and the influence of the operating environment. For example, the modulation effect of dust concentration change on vibration amplitude is analyzed, or the combined influence of humidity on electrical insulation and vibration response is considered. The key to feature extraction is to deeply integrate time-domain and frequency-domain features with multi-modal interaction characteristics through algorithm modeling and quantification to form a complete set of high-dimensional feature vectors, which not only retains the key information of a single signal but also captures the complex correlation between multiple signals, laying a solid data foundation for subsequent modeling and analysis.

[0035] Step S300: Perform graph construction processing according to the set of high-dimensional feature vectors. By analyzing the physical correlation between different features and the actual influence degree of the operating environment, a set of feature correlation graphs is obtained.

[0036] It is understandable that each feature in the high-dimensional feature vector set is represented as a node of a graph, and the edges between the nodes reflect the correlation strength and mutual influence law between the features. The graph construction process first models the initial connection relationship between the nodes based on physical correlation, such as the strong coupling relationship between the high-frequency energy of the vibration signal and the amplitude of the rotational speed fluctuation, or the modulation effect of the increase in humidity on the harmonic distortion in the electrical signal. Subsequently, by analyzing the actual impact of environmental parameters (such as dust concentration and temperature) on the operating state of the bucket wheel stacker-reclaimer, the weights of the node connection edges are dynamically adjusted. For example, an increase in dust concentration may strengthen the influence weight on the vibration characteristics of the main bearing, while a high-temperature environment may weaken the feature coupling of the rotational speed signal. In addition, a global structure optimization strategy is introduced during the construction process to ensure that the feature association graph can effectively reflect the dynamic feature interaction patterns of the bucket wheel stacker-reclaimer under different working conditions.

[0037] Step S400: Perform embedding processing on the feature association graph set based on a preset graph neural network, and generate a set of fused feature vectors through graph convolution operations and feature aggregation processing;

[0038] In this step, through graph convolution operations, the node information in the feature association graph is interactively aggregated with the information of neighboring nodes to capture the local and global dynamic relationships between each node. The graph convolution operation first aggregates the features within the node neighborhood layer by layer based on the structure of the feature association graph. For example, the high-frequency energy feature of the vibration signal node is aggregated with the dynamic change feature of the rotational speed node during the convolution process, and at the same time, the modulation effect of the dust concentration node on the two is combined, so as to form a more profound feature expression of the operating state. In order to enhance the interactive expression of environmental parameters and signal features, the embedding processing combines the time series features (such as the time-varying characteristics of rotational speed fluctuations) with the static conditions of environmental parameters (such as temperature and humidity), and gradually expands the perception range of each node through multiple layers of graph convolution, so that the features of distant neighboring nodes can also be incorporated into the final embedding representation. At the same time, feature aggregation further integrates the results of each layer of convolution through the weighted fusion of the features of the node and its neighboring nodes to form a node embedding vector with both local characteristics and global structure. Finally, the generated set of fused feature vectors can not only efficiently represent the complex associations of multi-modal signals, but also adapt to the dynamic change characteristics of the bucket wheel stacker-reclaimer under various working conditions.

[0039] Step S500: Perform modeling processing based on the set of fused feature vectors. By performing clustering analysis on the fused feature vectors of different operating conditions, an anomaly detection benchmark model is constructed. The anomaly detection benchmark model includes the feature center and standard deviation range of each operating condition;

[0040] Specifically, first, each vector in the fused feature vector set represents the comprehensive expression of multi-modal features during a certain operation cycle of the bucket wheel stacker-reclaimer. Combining the spatio-temporal dynamic correlation of vibration, rotational speed, electrical signals, and environmental parameters fully describes the operating state of the equipment under complex working conditions. Through cluster analysis, the fused feature vectors are divided into different working condition categories (such as startup, stable operation, commutation, etc.) according to their inherent similarity, and the central value and standard deviation range of the feature distribution are extracted for each working condition category. In the high-dimensional feature space, the features are classified according to the density distribution and distance threshold between vectors to ensure that the model can adapt to the complexity of working conditions and the non-linear characteristics of signals. For example, in the startup working condition, the vibration characteristics may be concentrated in the low-frequency region with high amplitudes, while in the stable operation working condition, the vibration amplitudes may be low and the main frequency distribution is stable. Extracting these distribution rules through clustering helps to accurately describe the operating characteristics of different working conditions. In addition, the extraction of the standard deviation range is used to construct the feature fluctuation boundary for each working condition, so as to distinguish the normal state and the abnormal state in subsequent real-time detection.

[0041] Step S600: Obtain the real-time data of the bucket wheel stacker-reclaimer, and input the real-time data into the anomaly detection reference model for comparison. By detecting the deviation degree of real-time features, an anomaly detection result is obtained, and the anomaly detection result includes whether there is an anomaly and the anomaly classification result.

[0042] In specific implementation, the real-time data is first processed by feature extraction to generate a high-dimensional vector representation consistent with the fused feature vector set, ensuring that the input data can match the feature center and standard deviation range of the anomaly detection reference model. During the comparison process, by calculating the deviation degree between the real-time features and the feature center of the corresponding working condition in the reference model, for example, quantifying the difference between the real-time features and the normal distribution through the Mahalanobis distance or Euclidean distance. When the deviation value exceeds the normal range defined by the reference model, it is determined as an anomaly. In addition, the deviation patterns of different features can also be used to classify the anomaly types. For example, the deviation of the frequency components of the vibration signal may indicate a mechanical fault, while the harmonic distortion of the electrical signal exceeding the range may point to an electrical system problem. Through the classification result, the source of the anomaly and the faulty part of the equipment can be further refined.

[0043] Furthermore, step S200 includes steps S210 to S240.

[0044] Step S210: Perform time-domain and frequency-domain feature extraction processing on the vibration signals of each cycle in the historical dataset. Extract time-domain features by calculating the mean value, kurtosis, and peak factor of the vibration signals, and extract frequency-domain features by calculating the main frequency and high-frequency energy ratio through fast Fourier transform to obtain a vibration feature matrix;

[0045] First, time-domain feature extraction mainly focuses on the amplitude variation and statistical characteristics of vibration signals, including the mean value (reflecting the overall level of the vibration signal and used to judge the stability of equipment operation), kurtosis (measuring the peak degree of the signal and used to capture mechanical impact faults, such as bearing or gear faults), and peak factor (revealing the abnormal amplitude of the signal through the ratio of the maximum amplitude to the root mean square value). These time-domain features can directly reflect the operating state of mechanical components, especially the vibration characteristics differences under conditions such as commutation and startup.

[0046] The extraction of frequency-domain features relies on the fast Fourier transform (FFT) to convert the vibration signal from the time domain to the frequency domain for identifying the mechanical characteristics corresponding to specific frequency components. The extraction of the main frequency can capture the main vibration modes of the equipment, such as the rotation frequency of the main shaft or the gear meshing frequency; the high-frequency energy ratio is used to detect the energy proportion of the high-frequency part in the vibration signal, which is crucial for capturing high-frequency features such as bearing wear or cracks. By combining time-domain and frequency-domain features, the characteristics of the vibration signal can be described from multiple dimensions.

[0047] The finally obtained vibration feature matrix, where each row in the matrix represents a set of features within an operating cycle, covering time-domain and frequency-domain information. This step realizes the comprehensive characterization of the equipment operating state by efficiently extracting the core features of the vibration signal of the bucket wheel stacker-reclaimer, especially the sensitive detection ability for early mechanical faults, and at the same time provides a high-quality feature basis for the subsequent fusion with multi-modal features such as rotational speed and electrical signals.

[0048] Step S220: Perform dynamic feature extraction processing based on the rotational speed signals of each cycle in the historical dataset, and obtain the rotational speed feature vector by calculating the fluctuation amplitude and the limit rotational speeds at startup and stop for the rotational speed signals;

[0049] It can be understood that the calculation of the fluctuation amplitude is used to quantify the change range of the rotational speed signal within a cycle, reflecting the stability of equipment operation or the load change situation. For example, under stable operating conditions, the fluctuation amplitude should be small, while under conditions such as commutation or overload operation, the fluctuation amplitude may increase significantly, indicating the instability of the mechanical or power system. Secondly, the limit rotational speeds at startup and stop directly reveal the power limit performance of the equipment by extracting the maximum and minimum values of the rotational speed signal during the startup and stop phases. This feature is particularly suitable for judging the operating ability of the drive motor or transmission system under high loads. For example, a low limit rotational speed may indicate insufficient motor output or the existence of mechanical resistance.

[0050] By extracting these dynamic features, a rotational speed feature vector is generated, where the feature information of each cycle is encoded as a high-dimensional vector. This vector not only includes the global operating stability of the equipment (such as the fluctuation amplitude), but also includes the dynamic performance of the equipment under key operating conditions (such as startup and shutdown). The rotational speed feature vector helps to identify faults such as overload, mechanical jamming, or abnormal power output by capturing the dynamic change characteristics of the rotational speed, thus enhancing the comprehensive monitoring ability of the bucket wheel stacker-reclaimer's operating state.

[0051] Step S230: Perform feature extraction processing based on the electrical signals and environmental parameters of each cycle in the historical dataset. Calculate the harmonic distortion rate by applying the fast Fourier transform to the electrical signals, and calculate the environmental impact factor using the temperature, humidity, and dust concentration in the environmental parameters to obtain the electrical-environmental feature matrix.

[0052] First, the feature extraction of the electrical signals calculates the harmonic distortion rate through the fast Fourier transform (FFT). The harmonic distortion rate can quantitatively reflect the degree to which the current or voltage signal in the motor drive system deviates from the sine wave form. A higher harmonic distortion rate usually indicates problems in the electrical system, such as inverter failures, harmonic interference caused by non-linear loads, or abnormal loads during equipment operation. This feature of the electrical signals plays a key role in judging the working efficiency and stability of the motor, especially in the case of overload or unstable operation conditions.

[0053] Second, the feature extraction of the environmental parameters (such as temperature, humidity, dust concentration) focuses on calculating the environmental impact factor to quantify the actual impact of external conditions on the equipment operating state. The calculation of the environmental impact factor can be normalized and weighted through a model. For example, high humidity reduces the lubrication effect of the main bearing, high dust concentration exacerbates the wear of the equipment surface or blocks mechanical components, and high temperature conditions impose additional pressure on the motor load. By constructing the environmental factor, these complex external conditions can be mapped into a comprehensive index to reflect the degree of influence of the environment on the equipment operating stability and health status.

[0054] Finally, the extracted electrical signal features and environmental parameter features are integrated into the electrical-environmental feature matrix. Each row of the matrix corresponds to the electrical features (such as harmonic distortion rate) and environmental features (such as environmental impact factor) of one cycle. This feature matrix realizes the deep coupling of electrical signals and environmental parameters, and can not only capture the dynamic characteristics within the electrical system, but also reflect the potential impact of external environmental changes on the operating state of the bucket wheel stacker-reclaimer. This multi-modal feature fusion provides comprehensive input data for subsequent feature correlation analysis and anomaly detection, thus significantly enhancing the operating monitoring ability of the bucket wheel stacker-reclaimer under complex working conditions and extreme environments.

[0055] Step S240: Integrate the vibration feature matrix, the rotational speed feature vector, and the electrical-environment feature matrix to obtain a high-dimensional feature vector set.

[0056] For each row of the finally generated high-dimensional feature vector set, it represents the multi-modal feature expression of the bucket wheel stacker-reclaimer within one operating cycle, which is a comprehensive integration of the time domain, frequency domain, dynamic characteristics, and environmental impacts.

[0057] Furthermore, step S300 includes steps S310 to S330.

[0058] Step S310: Initialize the processing according to the high-dimensional feature vector set diagram. By analyzing the physical correlations among the vibration signal, rotational speed signal, electrical signal, and environmental parameters of the bucket wheel stacker-reclaimer, and combining the influence of the rotational motion of the bucket wheel stacker-reclaimer on the vibration of the main bearing and the coupling characteristics between the motor load and rotational speed, construct a set of node relationship diagrams reflecting the interactions between components.

[0059] In this process, different features in the high-dimensional feature vector (such as vibration features, rotational speed features, electrical features, and environmental parameters) are mapped to nodes, and the edges between the nodes represent their physical correlation strength or interaction rules. For example, the edge between the vibration signal node and the rotational speed signal node can reflect the modulation effect of rotational speed fluctuations on the vibration amplitude and frequency, while the connection between the electrical signal node and the environmental parameter node characterizes the influence of environmental conditions (such as temperature and humidity) on the motor load and harmonic distortion.

[0060] During the process of graph initialization, it is also necessary to analyze in combination with the actual working characteristics of the bucket wheel stacker-reclaimer. Specifically, the rotational motion of the bucket wheel stacker-reclaimer usually has a direct vibration impact on the main bearing. Therefore, the correlation strength between the main bearing vibration signal node and the rotational speed node needs to be dynamically adjusted according to the fluctuation amplitude of the rotational speed; and the coupling relationship between the motor load and rotational speed also needs to extract the influence mode of current change on rotational speed stability through historical data analysis. In addition, environmental parameters (such as dust concentration) have an indirect impact on multiple signal nodes. For example, dust will accelerate the wear of the main bearing and increase the vibration amplitude, and may also have an additional impact on the motor load. Therefore, their correlations need to be reflected through global modeling.

[0061] Through the graph initialization process, the constructed set of node relationship diagrams can clearly express the interactions among various components and features during the operation of the bucket wheel stacker-reclaimer, providing a clear structured input for subsequent graph optimization and embedding processing.

[0062] Step S320: Perform environmental impact weight fusion processing according to the node relationship diagram. By introducing changes in dust concentration in the simulation environment to analyze the dynamic effects of dust on the operating states of the components of the bucket wheel stacker-reclaimer, and combining the response of the vibration signal to rotational speed changes, the influence of humidity on the lubrication state of the main bearing, and the influence of temperature on the motor load, modify the edge weights between the nodes to obtain a set of corrected feature diagrams.

[0063] It should be noted that, first of all, the influence of environmental parameters (such as dust concentration, humidity, temperature) on the operating state of the bucket wheel stacker-reclaimer is multi-dimensional and dynamic. It is necessary to analyze the action intensity of different environmental factors on each characteristic node under specific operating conditions through a simulation environment. For example, the change in dust concentration directly affects the wear degree of the main bearing, thereby causing an increase in the amplitude of the vibration signal. At the same time, dust has an indirect impact on the speed signal. For example, due to the increase in mechanical resistance caused by dust accumulation, speed fluctuations occur. Similarly, the influence of humidity on the lubrication state of the main bearing leads to a decrease in lubrication effect, and then an increase in the high-frequency components in the vibration signal; while the increase in temperature increases the motor load, thus affecting the current fluctuation and harmonic distortion characteristics.

[0064] In specific implementation, by analyzing the simulation environment data and historical operation data, the influence of these environmental factors on the edge weights between nodes is quantified. For example, the edge weight between the dust concentration and the vibration node is dynamically corrected, and the correction value is calculated according to the change rate of the dust concentration and its non-linear influence function on the main bearing vibration; the influence intensity of humidity on the lubrication state is determined by the amplitude response of the change in the vibration frequency of the main bearing; and the correction of the edge weight of the temperature on the electrical node can be dynamically calculated based on the coupling relationship between the dynamic load change of the current signal and the temperature.

[0065] Finally, the set of corrected feature maps generated through the fusion processing of environmental influence weights can comprehensively and dynamically reflect the modulation effect of environmental conditions on the operating state of the bucket wheel stacker-reclaimer and the correlation between various signal characteristics.

[0066] Step S330: Perform global structure optimization processing according to the corrected feature map. By analyzing the change in the interaction relationship between nodes under different operating conditions of the bucket wheel stacker-reclaimer, and combining the dynamic effects between nodes during the startup, stable operation, and commutation processes of the bucket wheel stacker-reclaimer, adjust the connection structure and edge weights of the nodes in the feature map to generate a set of feature correlation maps.

[0067] It can be understood that under different operating conditions, the interaction relationship between nodes will show significant dynamic changes. For example, in the startup condition, the interaction relationship between the speed node and the vibration node is relatively strong because the rapid increase in speed will directly cause a drastic change in the vibration characteristics; while in the stable operation condition, the interaction between the vibration and speed nodes tends to be stable, and instead, the coupling between the electrical node and the environmental node (such as the influence of temperature on current) becomes the main feature.

[0068] The key to global structure optimization is to adjust the connection structure between nodes and the edge weights according to the dynamic change patterns under different working conditions, so that the feature correlation graph can not only reflect the main features of the current operating condition, but also ensure the consistency and coherence of the relationships between different nodes through global constraints. For example, by analyzing the changing trend of the edge weights between the vibration node and the rotational speed node under different working conditions, the connection strength during startup and commutation conditions is dynamically strengthened, while this association is appropriately weakened during stable operating conditions. In addition, for the connection between the environmental node and other nodes, the optimization process dynamically adjusts its influence intensity on the vibration or electrical nodes according to the working condition classification to better fit the actual working conditions. Through global structure optimization, the node connection structure and edge weights in the feature correlation graph set can dynamically adapt to various operating states of the bucket wheel stacker-reclaimer, highlighting the main interaction relationships and weakening redundant or weakly associated connections.

[0069] Further, step S330 includes steps S331 to S333.

[0070] Step S331: Conduct dynamic interaction analysis and processing between nodes based on the corrected feature graph. By combining the time series pattern matching of vibration signals and rotational speed signals, the correlation analysis of electrical signals and environmental parameters, and using the dynamic time warping algorithm to extract the interaction change characteristics between nodes, a node interaction intensity matrix is generated;

[0071] The calculation formula for the interaction intensity between nodes is:

[0072]

[0073] where, W ij represents the interaction intensity between node i and node j; N represents the total number of time series involved in the calculation; T represents the number of sampling time points; t represents the serial number of the time point; represents the feature value of node i at time point t; represents the feature value of node j at time point t; represents the dynamic time warping distance between the time series features of node i and node j; σ ij represents the normalization factor of the dynamic time warping distance between node i and node j; ρ ij represents the feature weight ratio between node i and node j; Corr(X i , X j ) represents the Pearson correlation coefficient of the global time series between node i and node j.

[0074] Through this step, the generated node interaction intensity matrix W ijQuantify the dynamic interaction characteristics between multi-modal signals, capture both the local pattern matching characteristics of time series and reflect the global correlation, and solve the problem that it is difficult to quantify the complex interaction relationship between multiple signals in traditional methods. This matrix provides accurate input data for subsequent node optimization and edge weight adjustment, and at the same time enhances the adaptability and accuracy of the feature correlation graph to complex working conditions.

[0075] Step S332: Perform optimization processing according to the node interaction intensity matrix. Screen the edge connections with high interaction intensity through a sparsification algorithm, remove redundant node relationships, and strengthen important connections by combining the multi-node cooperation law of the bucket wheel stacker-reclaimer under the commutation working condition to generate an optimized node connection graph.

[0076] It should be noted that in the node interaction intensity matrix, the interaction intensity between each node represents the strength of their interaction. However, due to the complex multi-modal features and the number of nodes, the matrix contains a large number of low-intensity edges, which may be noise or weakly correlated redundant information. Therefore, optimization is required through a sparsification algorithm.

[0077] First of all, the sparsification process retains the high-intensity edges with an interaction intensity greater than the set interaction intensity threshold, and deletes the weak edge connections below the threshold at the same time. This process can effectively reduce the complexity of the connection graph and highlight the main interaction relationships between strongly correlated nodes. For example, in the nodes of vibration signals and rotational speed signals, only those edges that show significant coupling under the commutation working condition will be retained, while those with weak correlations will be removed.

[0078] Secondly, for the multi-node cooperation of the bucket wheel stacker-reclaimer under the commutation working condition, the optimization process further strengthens the key interaction connections between nodes. In the commutation working condition, vibration signals, rotational speed signals, and electrical signals often form strong dynamic linkage relationships. For example, a sudden change in rotational speed will cause a sharp fluctuation in the amplitude of the vibration signal, and at the same time, the current fluctuation may also increase significantly. During the optimization process, a collaborative analysis method is used to dynamically adjust the edge weights according to the changing trend of node characteristics under the commutation working condition. For example, the weight between the rotational speed node and the vibration node is increased to reflect their higher cooperation. Finally, the optimized node connection graph not only significantly simplifies the structure of the feature graph, retains the key nodes and high-intensity edges, but also can highlight the main interaction patterns under different working conditions.

[0079] Step S333: Perform edge weight adjustment processing according to the optimized node connection graph. By combining the node interaction law of the working condition category and the overall operating state, use the global regularization distribution optimization algorithm to adjust the edge weight distribution, match the connection relationship between nodes with the overall operating characteristics of the bucket wheel stacker-reclaimer, and generate a set of feature correlation graphs.

[0080] It is understandable that, first, by analyzing the node interaction characteristics of the bucket wheel stacker-reclaimer under typical working conditions (such as start-up, stable operation, and commutation), the edge weight adjustment rules under each working condition are extracted. For example, in the start-up working condition, the interaction between the vibration signal and the rotational speed signal shows a strong coupling relationship. Therefore, the edge weight between them needs to be weighted and increased to better reflect the dynamic response characteristics of the mechanical system. Under the stable operation working condition, the modulation effect of environmental parameters (such as temperature and humidity) on electrical signals is more significant. Therefore, it is necessary to appropriately increase the edge weight between electrical nodes and environmental nodes. Second, the global regularization distribution optimization algorithm is used to dynamically adjust the edge weights. The goal of the optimization algorithm is to ensure that the edge weight distribution can reflect the interaction intensity of local nodes through global constraints and maintain the topological structure consistency of the overall feature association graph. Specifically, the global regularization normalizes the edge weights and adjusts the edge weights in combination with the connection centrality and local density of the nodes. For example, if some nodes are key nodes under multiple working conditions (such as vibration nodes and rotational speed nodes), the adjustment of their edge weights needs to consider their global role in the overall graph structure. In addition, the dynamic adjustment of edge weights also takes into account the heterogeneous relationship between different signal types. For example, the time series coupling of rotational speed signals and vibration signals needs to be combined with the static conditions of environmental parameters, and the optimization results reflect the differences in such multimodal characteristics. The finally generated set of feature association graphs can accurately reflect the feature interaction patterns and dynamic association characteristics of the bucket wheel stacker-reclaimer under different operating conditions. The calculation formula for the edge weight is:

[0081]

[0082] where, represents the optimized edge weight; w ij represents the initial edge weight between node i and node j before optimization; w ik represents the initial edge weight between node i and node k; λ1, λ2, and λ3 are weight parameters. λ1 is used to control the consistency between the optimization result and the initial edge weight, λ2 is used to control the contribution of the influence of neighbor nodes to the edge weight, and λ3 is used to control the influence of the global regularization distribution on the edge weight optimization; D ij represents the distance between node i and node j; σ represents the regularization coefficient of the distance; N(i) represents the set of neighbor nodes of node i; k represents the serial number of the neighbor node; |N(i)| represents the number of neighbor nodes.

[0083] Furthermore, step S400 includes steps S410 to S430.

[0084] Step S410: Initialize the node embedding according to the set of feature association graphs. By mapping the node attributes in the feature association graph to low-dimensional embedding vectors and combining the weights of the edges in the graph to assign the initial interaction relationship strength, a set of node embedding graphs is obtained;

[0085] It can be understood that the specific implementation includes two key steps. First, for node attribute embedding, an embedding mapping function (such as linear dimensionality reduction or non - linear embedding algorithm) is used to convert high - dimensional feature vectors into low - dimensional representations. For example, features such as the high - frequency energy ratio and peak factor of a vibration node are encoded into a low - dimensional vector, which can compactly represent the main characteristics of the node while reducing the computational complexity. Second, the interaction intensity between nodes is initialized by combining the edge weights in the graph structure. The embedding vector of each node is determined not only by its own features but also affected by the edge weights of its neighbor nodes. For example, if the edge weight between the rotational speed node and the vibration node is high, the initial embedding vector of the vibration node will incorporate the attributes of the rotational speed node to a greater extent, thus reflecting their strong interaction relationship.

[0086] Step S420: Perform graph convolution on the set of node - embedded graphs. Each node aggregates the embedded features from its neighbor nodes, and combines the influence of rotational speed fluctuations on vibration nodes, the modulation effect of humidity on the lubrication state of the main bearing, and the interference intensity of dust concentration on electrical loads in the operation state of the bucket wheel stacker - reclaimer to dynamically adjust the edge weights, obtaining an updated set of node - embedded graphs;

[0087] During the graph convolution process, the embedding vector of each node not only contains its own features but also fuses the dynamic information of other nodes through the interaction relationships with neighbor nodes. For example, the embedding vector of the vibration node will comprehensively consider the change characteristics of the rotational speed node and the modulation effect of the humidity node on the lubrication state. The dynamic adjustment of edge weights further strengthens the physical association between features: the significant influence of rotational speed fluctuations on the vibration node requires an increased weight value during startup or commutation conditions, the modulation effect of humidity on the vibration of the main bearing enhances the connection strength between the humidity node and the vibration node through abnormal lubrication states, and the interference intensity of dust concentration on electrical nodes (such as current fluctuations and harmonic distortion) dynamically adjusts the edge weights according to the dust level. Through multiple convolution iterations, the embedding vector of each node is updated to contain both its own features and the spatio - temporal interaction information of neighbor features. For example, the updated vector of the vibration node combines the dynamic changes of the rotational speed node and the modulation features of the environmental node. Finally, the updated set of node - embedded graphs more accurately reflects the multi - modal dynamic interaction characteristics of the bucket wheel stacker - reclaimer under complex operating conditions, overcomes the defect that traditional static graph models are difficult to adapt to dynamic conditions, provides a more robust and accurate structured feature input for subsequent analysis, and explicitly models the dynamic influence of environmental parameters on signal features, significantly enhancing the applicability and diagnostic ability of the feature - correlation graph in actual operating conditions.

[0088] Step S430: Perform multi-layer graph convolution processing based on the updated node embedding graph set. By combining the time series features of signal nodes with the spatial state information of environmental parameters, construct a feature mapping mechanism based on time and space interaction. In each layer of convolution, co-update the dynamic change features of signal nodes and the static conditions of environmental parameters, and map all node embedding vectors into numerical representations to obtain a set of fused feature vectors.

[0089] It should be noted that in each layer of graph convolution, the dynamic features of signal nodes will be combined with the static states of environmental parameter nodes. By aggregating the embedding vectors of neighbor nodes, the multi-modal correlation characteristics under specific working conditions are dynamically captured. For example, when the humidity increases, the modulation effect of the humidity node on the vibration signal is manifested as enhanced high-frequency features. When the dust concentration is relatively high, the interference of the dust node on the electrical signal causes the harmonic distortion rate to rise. These relationships will be reflected in the node update through edge weight adjustment during the convolution process.

[0090] Through gradually expanding the perception range of nodes layer by layer, multi-layer graph convolution not only aggregates the features of direct neighbor nodes but also gradually introduces the influence of distant neighbor nodes, enabling the final node embedding vectors to reflect the collaborative relationships of global features. For example, the vibration characteristics of the main bearing will be combined with the dynamic change characteristics of the rotational speed node in one layer of convolution. After multi-layer convolution, indirect associations with electrical nodes and environmental nodes may also be introduced, thus more comprehensively reflecting the operating state of the equipment. In addition, the time series features and spatial state information are jointly updated during the convolution process. The dynamic change features of the time series and the static conditions of the environment are synchronously embedded into the node representations, ensuring that the model can generate consistent and comprehensive feature expressions in dynamic working conditions and complex environments. Finally, all node embedding vectors are mapped into numerical representations and integrated into a set of fused feature vectors. This set not only retains the local characteristics of the nodes but also captures the global graph structure and spatio-temporal interaction features.

[0091] Furthermore, step S500 includes steps S510 to S530.

[0092] Step S510: Perform clustering processing based on the set of fused feature vectors. Use the density clustering algorithm to group the high-dimensional features of signal nodes and environmental nodes, and combine the characteristic features of the bucket wheel stacker-reclaimer during startup, constant speed, and commutation operation conditions to divide the feature samples into different working condition categories to obtain a feature clustering result with working condition labels.

[0093] It is understandable that the density clustering algorithm can identify high-density regions in the set of fused feature vectors by defining the density relationship between sample points, group similar feature samples into the same class, and effectively isolate low-density abnormal samples. This feature is particularly applicable to the complex distribution of multi-modal features of the bucket wheel stacker-reclaimer. For example, under the starting condition, the high-frequency features of the vibration signal and the dynamic changes in the rotational speed show strong coupling, while in the uniform speed condition, these features are concentrated in the low-amplitude range. Through the density clustering algorithm, the feature samples under the starting condition will naturally form high-density clusters, while the samples under the uniform speed condition and the reversing condition are respectively aggregated in different feature space regions.

[0094] Secondly, the clustering process combines the physical characteristics of different working conditions of the bucket wheel stacker-reclaimer to ensure the accuracy of the clustering results. For example, the feature samples under the starting condition usually show high vibration amplitude, high rotational speed change rate, and strong influence of environmental parameters; the samples under the uniform speed condition are mainly concentrated in the lower feature fluctuation range; in the reversing condition, due to the sudden change in rotational speed and the dynamic fluctuation of vibration characteristics, the samples are distributed in a complex high-dimensional space. Through clustering, the feature patterns of these working condition characteristics can be effectively identified, and corresponding working condition labels can be assigned to each clustering cluster. This step solves the problem of identifying the distribution pattern of complex multi-modal data in a high-dimensional space through the density clustering algorithm, and significantly improves the physical interpretability of the clustering results by combining the actual working condition characteristics, providing a clear basis for working condition classification for subsequent working condition modeling and anomaly detection.

[0095] Step S520: Perform in-working-condition feature distribution analysis and processing based on the feature clustering results. By calculating the statistical range of the signal node features and environmental node features in each working condition category, a feature distribution model including parameters such as mean, variance, maximum value, and minimum value is formed.

[0096] Specifically, first, the feature clustering results provide a grouping of feature samples for each working condition category. For example, the starting condition includes the high-frequency amplitude feature of the vibration node, the high-fluctuation amplitude feature of the rotational speed node, and the dynamic influence characteristics of the humidity or dust node. By performing statistical analysis on the feature samples within these groupings, the distribution characteristics of each signal node and environmental node can be calculated, including the mean value of the feature (reflecting the typical level of the signal), variance (describing the fluctuation range of the signal), maximum value, and minimum value (indicating the limit range of the signal).

[0097] Secondly, the feature distribution analysis needs to be classified according to each node type and signal characteristic. For example, the mean value and variance of the vibration signal are used to characterize the overall stability of the main bearing vibration, while the maximum value and minimum value reflect the vibration limit amplitude under the starting or reversing condition; the distribution characteristics of the rotational speed signal can reflect the stability and fluctuation range of the power system under different working conditions; the statistical parameters of the environmental nodes (such as humidity or dust concentration) can reveal the long-term impact of the external environment on the equipment operation.

[0098] Step S530: Perform benchmark model construction processing according to the feature distribution model. By using the feature distribution parameters of each working condition category, establish a multi-dimensional Gaussian distribution model, define the reference boundaries for normal operation for the numerical ranges of signal nodes and environmental nodes, and integrate the data ranges of each working condition category to obtain an anomaly detection benchmark model.

[0099] It should be noted that the feature distribution model of each working condition category contains the distribution parameters of multiple signal nodes and environmental nodes, such as the mean and variance of vibration signals, the fluctuation range of rotational speed signals, and the statistical ranges of environmental parameters (such as humidity and dust concentration). These feature distributions are modeled through multi-dimensional Gaussian distributions to construct the probability density function of each working condition category.

[0100] In this distribution, the feature mean defines the central value of the feature, the variance describes the fluctuation range of the feature, and the covariance matrix of the multi-dimensional Gaussian distribution can also capture the correlation between different features. For example, the frequency distribution of vibration signals may be highly correlated with rotational speed fluctuations, and this correlation can be reflected by the covariance.

[0101] Specifically, for the feature set of a certain working condition category, its Gaussian distribution form is:

[0102]

[0103] where p(X) represents the probability density value of the multi-dimensional feature vector X, which is used to describe the possibility that this feature vector belongs to a certain working condition category; X is the feature vector, representing the set of multi-modal features of the bucket wheel machine under a certain working condition; n represents the dimension of the feature vector; μ is the mean vector, representing the expected values of each feature under the working condition category; Σ is the covariance matrix, representing the correlation and fluctuation range between features.

[0104] Based on the multi-dimensional Gaussian distribution model, the reference boundaries for normal operation within each working condition category can be defined. By setting thresholds (such as a 95% confidence interval), delimit the region where the feature values fall within the normal range. For example, for the vibration and rotational speed nodes in the startup working condition, if the probability density of a certain feature value is lower than the set threshold, it can be preliminarily judged that this feature is in an abnormal state. The introduction of the multi-dimensional Gaussian distribution model not only captures the statistical range of a single feature but also describes the cooperative relationship between multi-modal features, thus solving the problem that traditional single-variable threshold methods cannot handle the complexity of multi-modal data. In addition, by integrating the distribution information of multiple working condition categories, this benchmark model achieves a comprehensive adaptation to complex operating environments, can dynamically adjust the detection benchmark under multiple working conditions such as startup, stable operation, and commutation, thereby improving the robustness and application value of the bucket wheel machine anomaly detection and providing a scientific basis for real-time monitoring and intelligent maintenance.

[0105] Furthermore, step S600 includes steps S610 to S630.

[0106] Step S610: Perform feature extraction processing based on real-time data. By calculating the mean value, peak value, and fluctuation amplitude of the vibration signal within a short-time window, extract the current fluctuation range and change rate of the rotational speed signal, and extract the power factor and harmonic distortion degree of the electrical signal. At the same time, collect the real-time state values of environmental parameters and align the timestamps with the signal features, and integrate them to obtain a set of real-time feature vectors.

[0107] It can be understood that the feature extraction of the vibration signal calculates the mean value, peak value, and fluctuation amplitude through a short-time window. The mean value reflects the overall vibration level, the peak value captures transient impacts, and the fluctuation amplitude quantifies the dynamic range of vibration, which is used to judge the running stability of components such as main bearings and gearboxes. The rotational speed signal extracts the fluctuation range and change rate. The fluctuation range is used to monitor the smoothness of the power system, and the change rate reflects the dynamic adjustment ability under conditions such as starting and commuting. The electrical signal calculates the power factor and harmonic distortion degree. The power factor measures the operating efficiency of the motor, and the harmonic distortion degree extracts the non-linear fault features of the electrical system through Fourier transform, reflecting the health status of the electrical load. At the same time, the environmental parameters are collected in real-time and the timestamps are aligned with the signal features to ensure the temporal consistency of the data. Finally, these multi-modal features, including mechanical, power, electrical, and environmental parameters, are integrated into a set of real-time feature vectors to comprehensively reflect the operating state of the bucket wheel stacker-reclaimer.

[0108] Step S620: Perform deviation calculation processing based on the set of real-time feature vectors and the anomaly detection reference model. By matching the real-time operating condition category, compare the real-time features with the normal distribution parameters of the corresponding operating conditions in the reference model, and integrate the deviation values of all features to obtain a set of real-time deviation indicators.

[0109] It should be noted that in the benchmark model, each operating condition category contains multi-dimensional Gaussian distribution parameters such as the mean, variance, and covariance matrix of the feature distribution. These parameters define the numerical range and fluctuation law of the features under normal operating conditions. After the matching is completed, the real-time features are compared one by one with the distribution parameters of the corresponding operating condition category, and the deviation degree of each feature is quantified by calculating the deviation value. The calculation of the deviation value adopts the method based on Mahalanobis distance, which measures the weighted distance of the real-time feature relative to the distribution mean, and reflects the coupling relationship between features through the covariance matrix, so as to more accurately depict the overall deviation degree of multi-modal data. In the specific calculation, the deviation value not only includes individual features (such as the mean deviation of the vibration signal and the abnormal change of the rotational speed fluctuation), but also comprehensively considers the collaborative deviation of multiple features. For example, whether the dynamic coupling of the vibration and rotational speed signals in the starting condition conforms to the normal distribution range. At the same time, by integrating the deviation values of all features, a real-time deviation index set is generated. This set not only quantifies the abnormal degree of each feature, but also provides a comprehensive index of the overall deviation.

[0110] Step S630: Perform anomaly detection processing according to the real-time deviation index set. By comparing each deviation index with the anomaly threshold set in the benchmark model item by item, it is judged whether there is a feature deviation exceeding the threshold, and combined with the distribution range and source of the deviation feature, the anomaly detection result is output.

[0111] Specifically, the first step of anomaly detection is to compare these deviation values with the threshold set in the benchmark model one by one. The threshold is usually determined based on the feature distribution parameters of the operating condition category (such as the confidence interval of the multi-dimensional Gaussian distribution). For example, if the high-frequency energy deviation of the vibration signal exceeds the upper threshold of its corresponding operating condition, it indicates an abnormality of the mechanical component; similarly, an excessive deviation of the rotational speed change rate indicates an unstable power output. After the item-by-item comparison is completed, the anomaly detection further conducts a comprehensive analysis in combination with the distribution range and source of the deviation feature. For example, if multiple features (such as vibration amplitude and rotational speed fluctuation) simultaneously exceed the threshold and show coupled anomalies, it may point to a specific failure mode (such as gearbox wear or main bearing damage). In addition, abnormal deviations of environmental parameters (such as too high dust concentration) may also serve as external incentives, affecting the changes of other signal features. Through this multi-level analysis, the system can not only judge whether there is an anomaly, but also provide the anomaly classification result and the preliminary location of potential faults in combination with the specific source and distribution range of the deviation feature. This step realizes the accurate determination and anomaly classification of real-time feature deviations, and significantly improves the fault detection ability of the bucket wheel stacker-reclaimer under complex operating conditions.

[0112] Embodiment 2:

[0113] As Figure 2 shown, this embodiment provides a safety monitoring system for a bucket wheel stacker-reclaimer, and the system includes:

[0114] An acquisition module 901, configured to acquire a historical data set, where the historical data set includes vibration signals, rotation speed signals, electrical signals, and corresponding environmental parameters collected in each historical operation cycle of the bucket wheel stacker-reclaimer;

[0115] An extraction module 902, configured to perform feature extraction processing according to each cycle in the historical data set, and obtain a high-dimensional feature vector set by analyzing and quantifying the time domain, frequency domain, and multi-modal interaction features of various signals;

[0116] A construction module 903, configured to perform graph construction processing according to the high-dimensional feature vector set, and obtain a set of feature association graphs by analyzing the physical correlation between different features and the actual influence degree of the operating environment;

[0117] A fusion module 904, which performs embedding processing on the set of feature association graphs based on a preset graph neural network, and generates a set of fused feature vectors through graph convolution operations and feature aggregation processing;

[0118] A modeling module 905, configured to perform modeling processing according to the set of fused feature vectors, and construct an anomaly detection benchmark model by performing clustering analysis on the fused feature vectors of different operating conditions. The anomaly detection benchmark model includes the feature center and standard deviation range of each operating condition;

[0119] A detection module 906, configured to acquire real-time data of the bucket wheel stacker-reclaimer, and input the real-time data into the anomaly detection benchmark model for comparison, and obtain an anomaly detection result by detecting the deviation degree of real-time features. The anomaly detection result includes whether there is an anomaly and the anomaly classification result.

[0120] In a specific embodiment of the present invention, the extraction module 901 includes:

[0121] A first extraction unit, configured to perform time domain and frequency domain feature extraction processing according to the vibration signal of each cycle in the historical data set, extract time domain features by calculating the mean value, kurtosis, and peak factor of the vibration signal, and extract frequency domain features by calculating the main frequency and high-frequency energy ratio through fast Fourier transform, so as to obtain a vibration feature matrix;

[0122] A second extraction unit, configured to perform dynamic feature extraction processing according to the rotation speed signal of each cycle in the historical data set, and obtain a rotation speed feature vector by calculating the fluctuation amplitude and the limit rotation speed of starting and stopping of the rotation speed signal;

[0123] A third extraction unit is configured to perform feature extraction processing based on the electrical signals and environmental parameters in each cycle of the historical dataset. By applying fast Fourier transform to the electrical signals to calculate the harmonic distortion rate, and using the temperature, humidity, and dust concentration in the environmental parameters to calculate the environmental impact factor, an electrical-environmental feature matrix is obtained.

[0124] A first integration unit is configured to integrate the vibration feature matrix, the rotational speed feature vector, and the electrical-environmental feature matrix to obtain a high-dimensional feature vector set.

[0125] In a specific embodiment of the present invention, the construction module 902 includes:

[0126] A first construction unit is configured to perform initialization processing according to the high-dimensional feature vector set diagram. By analyzing the physical correlations of the vibration signals, rotational speed signals, electrical signals, and environmental parameters of the bucket wheel stacker-reclaimer, and combining the influence of the rotational motion of the bucket wheel stacker-reclaimer on the vibration of the main bearing and the coupling characteristics of the motor load and rotational speed, a set of node relationship diagrams reflecting the interaction between components is constructed.

[0127] A second construction unit is configured to perform environmental impact weight fusion processing according to the node relationship diagram. By introducing changes in dust concentration in the simulation environment to analyze the dynamic effects of dust on the operating states of the components of the bucket wheel stacker-reclaimer, and combining the response of the vibration signal to changes in rotational speed, the influence of humidity on the lubrication state of the main bearing, and the influence of temperature on the motor load, the edge weights between nodes are corrected to obtain a set of corrected feature diagrams.

[0128] A third construction unit is configured to perform global structure optimization processing according to the corrected feature diagram. By analyzing the changes in the interaction relationships of each node under different operating conditions of the bucket wheel stacker-reclaimer, and combining the dynamic effects between nodes during the startup, stable operation, and commutation processes of the bucket wheel stacker-reclaimer, the connection structure and edge weights of the nodes in the feature diagram are adjusted to generate a set of feature correlation diagrams.

[0129] Embodiment 3:

[0130] Corresponding to the above method embodiment, a bucket wheel stacker-reclaimer safety monitoring device is also provided in this embodiment. The bucket wheel stacker-reclaimer safety monitoring device described below can be mutually referred to with the bucket wheel stacker-reclaimer safety monitoring method described above.

[0131] Figure 3 is a block diagram of a bucket wheel stacker-reclaimer safety monitoring device 800 shown according to an exemplary embodiment. As Figure 3 shown, the bucket wheel stacker-reclaimer safety monitoring device 800 may include: a processor 801, a memory 802. The bucket wheel stacker-reclaimer safety monitoring device 800 may further include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0132] Among them, the processor 801 is used to control the overall operation of the bucket wheel stacker-reclaimer safety monitoring device 800 to complete all or part of the steps in the above-mentioned bucket wheel stacker-reclaimer safety monitoring method. The memory 802 is used to store various types of data to support the operation of the bucket wheel stacker-reclaimer safety monitoring device 800. These data may include, for example, instructions for any application or method operating on the bucket wheel stacker-reclaimer safety monitoring device 800, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, and so on. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. The multimedia component 803 may include a screen and an audio component. Among them, the screen can be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal can be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be a keyboard, a mouse, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the bucket wheel stacker-reclaimer safety monitoring device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, near field communication (NFC), 2G, 3G, or 4G, or a combination of one or more of them. Therefore, the corresponding communication component 805 may include: a Wi-Fi module, a Bluetooth module, and an NFC module.

[0133] In one exemplary embodiment, a bucket wheel stacker-reclaimer safety monitoring device 800 can be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components, and is used to execute the above-mentioned bucket wheel stacker-reclaimer safety monitoring method.

[0134] In another exemplary embodiment, a computer-readable storage medium including program instructions is further provided. When the program instructions are executed by a processor, the steps of the above-mentioned bucket wheel stacker-reclaimer safety monitoring method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions can be executed by a processor 801 of a bucket wheel stacker-reclaimer safety monitoring device 800 to complete the above-mentioned bucket wheel stacker-reclaimer safety monitoring method.

[0135] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention.

Claims

1. A bucket wheel machine safety monitoring method, characterized in that: include: Acquire a historical data set, wherein the historical data set includes a vibration signal, a speed signal, an electrical signal, and corresponding environmental parameters collected during each historical operation cycle of the bucket wheel machine; Performing feature extraction processing according to each cycle in the historical data set, and obtaining a high-dimensional feature vector set by analyzing and quantifying the time domain, frequency domain and multimodal interaction features of various types of signals; Performing graph construction processing according to the high-dimensional feature vector set, and obtaining a feature association graph set by analyzing the physical correlation between different features and the actual impact of the operating environment; The feature association graph set is embedded based on a preset graph neural network, and a fused feature vector set is generated through graph convolution operation and feature aggregation processing; Modeling is performed according to the fused feature vector set, and clustering analysis is performed on the fused feature vectors of different operating conditions to construct an anomaly detection benchmark model, wherein the anomaly detection benchmark model includes a feature center and a standard deviation range for each operating condition; Real-time data of the bucket wheel machine is obtained, and the real-time data is input into the anomaly detection benchmark model for comparison. By detecting the degree of deviation of the real-time features, an anomaly detection result is obtained, and the anomaly detection result includes whether there is an anomaly and an anomaly classification result.

2. A bucket wheel machine safety monitoring method according to claim 1, characterized in that: According to each period in the historical data set, feature extraction processing is performed, and by analyzing and quantifying the time domain, frequency domain and multimodal interaction features of various signals, a high-dimensional feature vector set is obtained, including: Performing time domain and frequency domain feature extraction processing on the vibration signal of each cycle in the historical data set, extracting time domain features by calculating the mean, kurtosis and peak factor of the vibration signal, and extracting frequency domain features by calculating the main frequency and high frequency energy ratio through fast Fourier transform, to obtain a vibration feature matrix; Performing dynamic feature extraction processing on the speed signal of each cycle in the historical data set, and obtaining a speed feature vector by calculating the fluctuation amplitude and the start-stop limit speed of the speed signal; Performing feature extraction processing on the electrical signal and environmental parameters of each cycle in the historical data set, calculating the harmonic distortion rate by applying fast Fourier transform to the electrical signal, and calculating the environmental impact factor by using the temperature, humidity and dust concentration in the environmental parameters to obtain an electrical-environmental feature matrix; The vibration characteristic matrix, the rotation speed characteristic vector and the electrical-environmental characteristic matrix are integrated to obtain a high-dimensional characteristic vector set.

3. A bucket wheel machine safety monitoring method according to claim 1, characterized in that: A graph construction process is performed according to the high-dimensional feature vector set, and a feature association graph set is obtained by analyzing the physical correlation between different features and the actual impact of the operating environment, including: According to the high-dimensional feature vector set graph initialization processing, by analyzing the physical association between bucket wheel vibration signal, speed signal, electrical signal and environmental parameters, combined with the influence of bucket wheel rotation on main bearing vibration and the coupling characteristics of motor load and speed, a node relationship graph set reflecting the interaction between components is constructed; Performing environmental impact weight fusion processing according to the node relationship graph, analyzing the dynamic effect of dust on the operating state of bucket wheel machine components by introducing dust concentration changes in the simulation environment, combining the response of vibration signals to speed changes, the influence of humidity on the lubrication state of the main bearing, and the influence of temperature on the motor load, correcting the edge weights between nodes to obtain a corrected feature graph set; A global structural optimization process is performed based on the modified characteristic graph. By analyzing the changes in the interaction relationship of each node under different operating conditions of the bucket wheel machine, combined with the dynamic effects between nodes during the startup, stable operation and reversing of the bucket wheel machine, the connection structure and edge weights of the nodes in the characteristic graph are adjusted to generate a set of characteristic association graphs.

4. A bucket wheel machine safety monitoring method according to claim 3, characterized in that: According to the modified characteristic graph, global structural optimization processing is performed, by analyzing the changes in the interaction relationship of each node under different operating conditions of the bucket wheel machine, combining the dynamic effects between nodes during the start-up, stable operation and reversing of the bucket wheel machine, adjusting the connection structure and edge weights of the nodes in the characteristic graph, and generating a characteristic association graph set, including: According to the modified characteristic graph, dynamic interaction analysis and processing between nodes is performed, and a node interaction strength matrix is ​​generated by combining time series pattern matching of vibration signals and speed signals, correlation analysis of electrical signals and environmental parameters, and extracting interaction change characteristics between nodes using a dynamic time warping algorithm; Optimization processing is performed according to the node interaction strength matrix, edge connections with high interaction strength are screened through a sparse algorithm, redundant node relationships are removed, and important connections are strengthened in combination with the multi-node synergy law of the bucket wheel excavator under the reversing condition to generate an optimized node connection diagram; The edge weight adjustment process is performed according to the optimized node connection graph. By combining the node interaction rules of the working condition category and the overall operating status, a global regularized distribution optimization algorithm is used to adjust the edge weight distribution, the connection relationship between the nodes is matched with the overall operating characteristics of the bucket wheel excavator, and a feature association graph set is generated.

5. A bucket wheel machine safety monitoring method according to claim 1, characterized in that: The feature association graph set is embedded based on a preset graph neural network, and a fused feature vector set is generated through graph convolution operation and feature aggregation processing, including: Performing node embedding initialization processing according to the feature association graph set, mapping node attributes in the feature association graph to low-dimensional embedding vectors, and assigning initial interaction relationship strengths in combination with edge weights in the graph, thereby obtaining a node embedding graph set; The node embedding graph set is subjected to graph convolution processing, and the edge weights are dynamically adjusted by aggregating embedding features of each node from its neighboring nodes, combining the influence of speed fluctuation on vibration nodes in the operation state of the bucket wheel machine, the modulation effect of humidity on the lubrication state of the main bearing, and the interference intensity of dust concentration on the electrical load, so as to obtain an updated node embedding graph set; Multi-layer graph convolution processing is performed according to the updated node embedding graph set. By combining the time series characteristics of the signal nodes and the spatial state information of the environmental parameters, a feature mapping mechanism based on the interaction of time and space is constructed. In each layer of convolution, the dynamic change characteristics of the signal nodes and the static conditions of the environmental parameters are collaboratively updated, and all node embedding vectors are mapped to numerical representations to obtain a fused feature vector set.

6. A bucket wheel machine safety monitoring method according to claim 1, characterized in that: Modeling is performed based on the fused feature vector set, and clustering analysis is performed on the fused feature vectors of different operating conditions to construct an anomaly detection benchmark model, including: Clustering is performed according to the fused feature vector set, high-dimensional features of signal nodes and environmental nodes are grouped using a density clustering algorithm, and feature samples are divided into different operating condition categories in combination with the characteristic characteristics of the bucket wheel machine's starting, uniform speed, and reversing operating conditions, to obtain feature clustering results with operating condition labels; Performing feature distribution analysis and processing within the working condition according to the feature clustering results, and forming a feature distribution model including mean, variance, maximum and minimum parameters by calculating the statistical range of signal node features and environmental node features in each working condition category; A benchmark model is constructed based on the characteristic distribution model. A multidimensional Gaussian distribution model is established by utilizing the characteristic distribution parameters of each operating condition category. The reference boundaries of normal operation are defined for the numerical ranges of signal nodes and environmental nodes. The data ranges of each operating condition category are integrated to obtain an anomaly detection benchmark model.

7. A bucket wheel machine safety monitoring method according to claim 1, characterized in that: The real-time data of the bucket wheel machine is obtained, and the real-time data is input into the anomaly detection benchmark model for comparison, and the anomaly detection result is obtained by detecting the degree of deviation of the real-time features, including: Perform feature extraction processing according to the real-time data, perform short-time window calculation on the mean, peak value and fluctuation amplitude of the vibration signal, extract the current fluctuation range and change rate of the speed signal, extract the power factor and harmonic distortion of the electrical signal, and collect the real-time state values ​​of the environmental parameters and align the timestamps with the signal features to obtain a real-time feature vector set through integration; Deviation calculation is performed according to the real-time feature vector set and the anomaly detection benchmark model, and the real-time features are compared with the normal distribution parameters of the corresponding working conditions in the benchmark model by matching the real-time working condition categories, and the deviation values ​​of all features are integrated to obtain a real-time deviation indicator set; Anomaly detection processing is performed based on the real-time deviation indicator set. By comparing the deviation indicators with the anomaly thresholds set by the benchmark model item by item, it is determined whether there are feature deviations exceeding the thresholds, and the anomaly detection results are output in combination with the distribution range and source of the deviation features.

8. A bucket wheel excavator safety monitoring system, characterized in that: include: An acquisition module is used to acquire a historical data set, wherein the historical data set includes a vibration signal, a speed signal, an electrical signal and corresponding environmental parameters collected in each historical operation cycle of the bucket wheel machine; An extraction module is used to perform feature extraction processing according to each period in the historical data set, and obtain a high-dimensional feature vector set by analyzing and quantifying the time domain, frequency domain and multimodal interaction features of various types of signals; A construction module, used to perform graph construction processing according to the high-dimensional feature vector set, and obtain a feature association graph set by analyzing the physical correlation between different features and the actual impact of the operating environment; A fusion module, which embeds the feature association graph set based on a preset graph neural network, and generates a fused feature vector set through graph convolution operation and feature aggregation processing; A modeling module, used for performing modeling processing according to the fused feature vector set, and constructing an anomaly detection benchmark model by clustering analysis of the fused feature vectors of different operating conditions, wherein the anomaly detection benchmark model includes a feature center and a standard deviation range for each operating condition; The detection module is used to obtain real-time data of the bucket wheel machine and input the real-time data into the anomaly detection benchmark model for comparison. By detecting the degree of deviation of the real-time features, an anomaly detection result is obtained. The anomaly detection result includes whether there is an anomaly and an anomaly classification result.

9. A bucket wheel excavator safety monitoring system according to claim 8, characterized in that: The extraction module comprises: A first extraction unit is used to perform time domain and frequency domain feature extraction processing on the vibration signal of each period in the historical data set, extract the time domain feature by calculating the mean, kurtosis and peak factor of the vibration signal, and extract the frequency domain feature by calculating the main frequency and high frequency energy ratio by fast Fourier transform to obtain a vibration feature matrix; A second extraction unit is used to perform dynamic feature extraction processing according to the speed signal of each period in the historical data set, and obtain a speed feature vector by calculating the fluctuation amplitude and the start-stop limit speed of the speed signal; a third extraction unit, configured to perform feature extraction processing according to the electrical signal and the environmental parameters of each cycle in the historical data set, calculate the harmonic distortion rate by applying fast Fourier transform to the electrical signal, and calculate the environmental impact factor by using the temperature, humidity and dust concentration in the environmental parameters to obtain an electrical-environmental feature matrix; The first integration unit is used to integrate the vibration characteristic matrix, the rotation speed characteristic vector and the electrical-environmental characteristic matrix to obtain a high-dimensional characteristic vector set.

10. A bucket wheel machine safety monitoring system according to claim 8, characterized in that: The building blocks include: A first construction unit is used to construct a node relationship graph set reflecting the interaction between components by analyzing the physical association between the vibration signal, speed signal, electrical signal and environmental parameters of the bucket wheel machine and combining the influence of the rotating motion of the bucket wheel machine on the vibration of the main bearing and the coupling characteristics of the motor load and the speed according to the initialization processing of the high-dimensional feature vector set graph; The second construction unit is used to perform environmental impact weight fusion processing according to the node relationship graph, analyze the dynamic effect of dust on the operating state of bucket wheel machine components by introducing dust concentration changes in the simulation environment, and combine the response of vibration signals to speed changes, the influence of humidity on the lubrication state of the main bearing, and the influence of temperature on the motor load to correct the edge weights between nodes to obtain a corrected feature graph set; The third construction unit is used to perform global structural optimization processing according to the modified characteristic graph, by analyzing the changes in the interaction relationship of each node under different operating conditions of the bucket wheel machine, combining the dynamic effects between nodes during the startup, stable operation and reversing of the bucket wheel machine, adjusting the connection structure and edge weights of the nodes in the characteristic graph, and generating a set of characteristic association graphs.

Images