Mining monorail crane control system and method based on modular integration

Abstract

The invention relates to the field of intelligent control, and discloses a mining monorail crane control system and method based on modular integration. The system comprises an operation state matrix generation module, an abnormal feature vector generation module, a driving part vibration spectrum generation module, a diagnostic element generation module, a traction dynamic distribution scheme and multi-union self-healing strategy generation module and a joint control signal generation module, and an operation state matrix is generated through time-space fusion based on monorail crane multi-source heterogeneous data; extracting a matrix abnormal feature vector, and outputting a driving part vibration spectrum through adversarial spectrum reconstruction; carrying out wavelet packet sample entropy decomposition by combining a load and a battery signal to generate a diagnosis element; decoupling a fault unit according to the fault rule base, and forming a traction dynamic distribution scheme and a multi-union self-healing strategy; and a joint control signal is generated and wirelessly transmitted to a main control core of the monorail crane. According to the invention, the problems of cascade paralysis caused by local faults and sudden shutdown caused by early weak fault leak detection of the mining monorail crane under the abnormal condition are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a control system and method for a mining monorail crane based on modular integration. Background Technology

[0002] When a critical power unit experiences a sudden malfunction, the existing mine monorail control system lacks a real-time energy flow topology reconstruction mechanism and dynamic load migration, making it unable to quickly isolate the faulty unit and replan the traction force transmission path. This leads to unstable system energy distribution, a surge in the risk of overload of the remaining healthy units, and in turn, triggers a chain reaction of multiple unit failures or even a complete collapse of the global traction function, thus inhibiting the safety of continuous transportation in the roadway and the reliability of equipment operation.

[0003] Existing fault detection technologies for mining monorail crane drive units are susceptible to interference from strong periodic background noise and operating condition fluctuations, making it difficult to effectively separate the early, weak damage characteristic signals hidden in the vibration spectrum. This results in insufficient sensitivity of traditional time-frequency analysis methods to capture nonlinear dynamic behavior, making it impossible to establish highly robust fault diagnosis units in complex operating environments. Consequently, the rate of missed early damage detection increases, predictive maintenance fails, and ultimately, the potential risks of sudden fractures of critical components and unplanned system downtime are significantly increased. Summary of the Invention

[0004] This invention provides a control system and method for mine monorail cranes based on modular integration. Its main purpose is to solve the problems of cascading fragility and low anti-interference ability in the control of mine monorail cranes based on modular integration.

[0005] To achieve the above objectives, this invention provides a modularly integrated control system for a mine monorail crane, characterized by comprising the following modules: an operating state matrix generation module, an abnormal feature vector generation module, a drive unit vibration spectrum generation module, a diagnostic element generation module, a traction force dynamic distribution scheme and multi-link self-healing strategy generation module, and a joint control signal generation module, wherein:

[0006] The operation status matrix generation module performs spatiotemporal fusion on the multi-source heterogeneous data of the monorail crane to obtain the operation status matrix of the monorail crane.

[0007] The abnormal feature vector generation module extracts fault features from the operating state matrix to obtain the abnormal feature vector of the monorail crane.

[0008] The drive unit vibration spectrum generation module performs adversarial spectrum reconstruction on the abnormal feature vector to obtain the drive unit vibration spectrum of the monorail crane.

[0009] The diagnostic element generation module performs wavelet packet sample entropy decomposition on the vibration spectrum of the drive unit based on the load weight signal and real-time battery status signal of the monorail, and obtains the diagnostic elements of the monorail.

[0010] The traction force dynamic distribution scheme and multi-link self-healing strategy generation module decouples the diagnostic primitives based on the fault rule base to obtain the traction force dynamic distribution scheme and multi-link self-healing strategy of the monorail.

[0011] The joint control signal generation module fuses the dynamic traction force allocation scheme with the multi-link self-healing strategy to obtain the joint control signal of the monorail main control core, and wirelessly transmits the joint control signal to the monorail main control core.

[0012] In a preferred embodiment, when the operation state matrix generation module performs spatiotemporal fusion of multi-source heterogeneous data from the monorail to obtain the operation state matrix of the monorail, it is specifically used for:

[0013] Trajectory interpolation is performed on the multi-source heterogeneous data of the monorail to obtain the equally spaced trajectory sequence of the monorail;

[0014] Envelope extraction is performed on the equally spaced trajectory sequence to obtain the state vector of the monorail crane;

[0015] The state vectors are stacked in a time sequence to obtain the operating state matrix of the monorail crane.

[0016] In a preferred embodiment, when the abnormal feature vector generation module performs fault feature extraction on the operating state matrix to obtain the abnormal feature vector of the monorail crane, it is specifically used for:

[0017] Discrete wavelet decomposition is performed on the operating state matrix to obtain the wavelet coefficient matrix of the monorail crane;

[0018] Singular value decomposition is performed on the wavelet coefficient matrix to obtain the amplitude domain feature vector of the monorail crane;

[0019] The variance threshold filtering of the amplitude feature vector is used to obtain the abnormal feature vector of the monorail crane.

[0020] In a preferred embodiment, when the abnormal feature vector generation module performs waveform characteristic quantization on the wavelet coefficient matrix to obtain the amplitude domain feature vector of the monorail crane, it is specifically used for:

[0021] The wavelet coefficient matrix is ​​dimensionally segmented to obtain the independent frequency band vibration matrix of the monorail crane;

[0022] The amplitude modulus of the independent frequency band vibration matrix is ​​taken to obtain the amplitude domain statistics of the monorail crane;

[0023] Based on the spectral kurtosis, the amplitude domain statistics are demodulated using the resonance band to obtain the amplitude domain feature sequence of the monorail crane;

[0024] Tensor folding is performed on the amplitude-domain feature sequence to obtain the amplitude-domain feature vector of the monorail crane.

[0025] In a preferred embodiment, when the drive unit vibration spectrum generation module performs adversarial spectrum reconstruction on the abnormal feature vector to obtain the drive unit vibration spectrum of the monorail crane, it is specifically used for:

[0026] The abnormal feature vector is subjected to windowed noise separation to obtain the background noise suppression signal of the monorail crane;

[0027] The phase-compensated correction is performed on the noise floor signal to obtain the phase-aligned spectrum of the monorail.

[0028] The vibration spectrum of the drive unit of the monorail is obtained by performing adversarial enhancement reconstruction on the phase-aligned spectrum.

[0029] In a preferred embodiment, when the diagnostic element generation module performs wavelet packet entropy decomposition on the vibration spectrum of the drive unit based on the load weight signal and real-time battery status signal of the monorail to obtain the diagnostic elements of the monorail, it is specifically used for:

[0030] Harmonic suppression is applied to the vibration signal of the drive unit to generate a baseline stable signal.

[0031] Based on the baseline stable signal, the load weight signal and the battery status signal are synchronously aligned to obtain the vibration data segment;

[0032] The vibration data segment is divided into frequency bands based on orthogonal wavelet basis to obtain the time-frequency sub-band signal of the monorail.

[0033] The time-frequency sub-band signal is recursively quantized based on phase space reconstruction to obtain the frequency band entropy value of the monorail.

[0034] The frequency band entropy values ​​are sorted in ascending order based on the frequency band number to obtain the complexity distribution feature vector of the monorail. The complexity distribution feature vector is then jointly encoded to obtain the diagnostic primitive of the monorail.

[0035] In a preferred embodiment, when the diagnostic primitive generation module performs recursive quantization of the time-frequency subband signal based on phase space reconstruction to obtain the frequency band entropy set of the monorail, it is specifically used for:

[0036] The time-frequency subband signal is encapsulated into a two-order vector to obtain the basic trajectory and extended trajectory of the monorail crane;

[0037] The basic trajectory and the extended trajectory are recursively scanned to obtain the dual similarity measure of the monorail crane;

[0038] Subband aggregation is performed on the dual similarity measure to obtain the frequency band entropy value set of the monorail crane.

[0039] In a preferred embodiment, when the traction force dynamic allocation scheme and multi-unit self-healing strategy generation module performs fault unit decoupling on the diagnostic primitives based on the fault rule base to obtain the traction force dynamic allocation scheme and multi-unit self-healing strategy for the monorail, it is specifically used for:

[0040] The diagnostic primitives are subjected to feature dimensionality reduction to obtain the independent fault index of the monorail crane;

[0041] The energy flow path is reconstructed for the set of independent fault units to obtain a dynamic traction force distribution view;

[0042] Multimodal fusion is performed on the dynamic traction force distribution view to obtain the adaptive distribution matrix of the monorail.

[0043] By redirecting the bus signal of the adaptive allocation matrix, the multi-link self-healing strategy of the monorail crane is obtained;

[0044] In a preferred embodiment, when the control signal generation module performs instruction fusion of the dynamic traction force allocation scheme and the multi-link self-healing strategy to obtain the control signal of the monorail main control core, and wirelessly transmits the control signal to the monorail main control core, it is specifically used for:

[0045] The traction dynamic distribution scheme and the multi-link self-healing strategy are timestamped to obtain a synchronous command stream.

[0046] The control field is filled into the synchronization command stream to obtain the joint control signal frame of the monorail crane;

[0047] The control signal frame of the monorail is modulated and transmitted to obtain the radio frequency signal of the monorail.

[0048] The radio frequency signal is spatially guided and propagated to obtain the main control core execution command of the monorail crane.

[0049] To address the above problems, the present invention also provides a control method for a mine monorail crane based on modular integration, the method comprising:

[0050] S1. Spatiotemporal fusion of multi-source heterogeneous data of the monorail crane is performed to obtain the operating state matrix of the monorail crane;

[0051] S2. Extract fault features from the operating state matrix to obtain the abnormal feature vector of the monorail crane;

[0052] S3. Perform adversarial spectrum reconstruction on the abnormal feature vector to obtain the vibration spectrum of the drive unit of the monorail crane;

[0053] S4. Based on the load weight signal and real-time battery status signal of the monorail, perform wavelet packet sample entropy decomposition on the vibration spectrum of the drive unit to obtain the diagnostic primitives of the monorail.

[0054] S5. Based on the fault rule base, the diagnostic primitives are decoupled from the fault units to obtain the dynamic traction force distribution scheme and multi-link self-healing strategy of the monorail.

[0055] S6. The dynamic traction force distribution scheme and the multi-link self-healing strategy are fused together to obtain the joint control signal of the monorail main control core, and the joint control signal is wirelessly transmitted to the monorail main control core.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. By decoupling fault units and reconstructing energy flow paths, the system generates dynamic traction force distribution schemes and multi-unit self-healing strategies in real time during local faults. Combined with bus signal redirection technology, it realizes fault unit isolation and healthy unit collaborative control, ensuring that the monorail can maintain stable traction force output under abnormal working conditions, thereby improving the system's operational resilience and safety.

[0058] 2. By employing adversarial spectrum reconstruction and wavelet packet sample entropy decomposition techniques, and fusing load weight and battery status signals, nonlinear dynamic features are extracted from the vibration spectrum to generate highly sensitive diagnostic primitives. This can accurately identify early and subtle faults in the drive unit, providing a reliable basis for predictive maintenance and reducing the risk of sudden downtime. Attached Figure Description

[0059] Figure 1 A system architecture diagram of a mine monorail control system based on modular integration provided in an embodiment of the present invention;

[0060] Figure 2 This is a flowchart illustrating a modularly integrated control method for a mining monorail crane, as provided in an embodiment of the present invention.

[0061] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0064] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0065] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0066] In practice, the server-side equipment deployed in a modularly integrated mining monorail crane control system may consist of one or more devices. This modularly integrated mining monorail crane control system can be implemented as: a business instance, a virtual machine, or hardware devices. For example, this modularly integrated mining monorail crane control system can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, this modularly integrated mining monorail crane control system can be understood as software deployed on a cloud node, used to provide a modularly integrated mining monorail crane control system to various user terminals. Alternatively, this modularly integrated mining monorail crane control system can also be implemented as a virtual machine deployed on one or more devices in a cloud node. This virtual machine contains application software for managing various user terminals. Alternatively, this modularly integrated mining monorail crane control system can also be implemented as a server composed of numerous identical or different types of hardware devices, with one or more hardware devices configured to provide a modularly integrated mining monorail crane control system to various user terminals.

[0067] In terms of implementation, the modularly integrated mine monorail crane control system and the user terminal are mutually compatible. That is, if the modularly integrated mine monorail crane control system is implemented as an application installed on a cloud service platform, then the user terminal is a client that establishes a communication connection with the application; or if the modularly integrated mine monorail crane control system is implemented as a website, then the user terminal is implemented as a webpage; or if the modularly integrated mine monorail crane control system is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0068] like Figure 1 The figure shown is a system architecture diagram of a modularly integrated mine monorail control system provided in an embodiment of the present invention.

[0069] The modularly integrated mining monorail crane control system 100 described in this invention can be housed in a cloud server. In terms of implementation, it can function as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed into a website. Depending on the functions implemented, the modularly integrated mining monorail crane control system 100 may include an operating state matrix generation module 101, an abnormal feature vector generation module 102, a drive unit vibration spectrum generation module 103, a diagnostic primitive generation module 104, a traction force dynamic distribution scheme and multi-link self-healing strategy generation module 105, and a joint control signal generation module 106. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, stored in the electronic device's memory.

[0070] In this embodiment of the invention, in the modularly integrated mine monorail crane control system, each of the above-mentioned modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. In the modularly integrated mine monorail crane control system provided by this embodiment of the invention, the applicable scope of the modularly integrated mine monorail crane control system architecture can be adjusted by adding modules and directly calling them without modifying the program code, achieving cluster-based horizontal expansion to quickly and flexibly expand the modularly integrated mine monorail crane control system. In practical applications, the above modules can be set in the same device or different devices, or in virtual devices, such as service instances in a cloud server.

[0071] The following describes, with reference to specific embodiments, each component and its specific workflow of the modularly integrated mine monorail control system:

[0072] In this embodiment of the invention, when the operation state matrix generation module performs spatiotemporal fusion of multi-source heterogeneous data from the monorail to obtain the operation state matrix of the monorail, it is specifically used for:

[0073] Trajectory interpolation is performed on the multi-source heterogeneous data of the monorail to obtain the equally spaced trajectory sequence of the monorail;

[0074] Envelope extraction is performed on the equally spaced trajectory sequence to obtain the state vector of the monorail crane;

[0075] The state vectors are stacked in a time sequence to obtain the operating state matrix of the monorail crane.

[0076] Specifically, the precise position coordinates of the monorail at user-specified time intervals are calculated using mathematical differences from the raw trajectory data collected during the monorail operation from different sensors, which have different timestamps, sampling frequencies, precision, and physical meanings. This ultimately generates a monorail spatial position sequence with strictly equal timestamps and continuous data points.

[0077] Specifically, based on the spatial position sequence of a monorail crane, the key feature extreme points or trend boundaries of these kinematic parameters within a specific time window are identified and captured by analyzing the curves of their core kinematic parameters changing over time. These extracted key feature points and statistics are then combined to form a structured, low-dimensional feature vector.

[0078] Specifically, these state vectors are arranged and combined in an ordered manner along the time dimension according to their corresponding timestamps to construct a multidimensional data matrix. The matrix's columns represent a discrete moment or a short time period in the time series, the columns represent each feature component in the state vector, and the elements are the numerical values ​​or identifiers of specific state features at a specific time point.

[0079] Furthermore, in time periods when the original data points are sparse, interpolation can fill in the information gaps and suppress high-frequency noise or abnormal fluctuations in the original data, generating denser trajectory points that better reflect the continuous motion process and are more in line with the laws of physical motion. This provides a common and reliable spatiotemporal reference for all subsequent modules that need to analyze the spatiotemporal behavior of monorail cranes.

[0080] Furthermore, the high-dimensional, dense raw trajectory data is compressed into low-dimensional, highly integrated feature vectors, reducing the computational and storage requirements for subsequent analysis and increasing the density of key information. Key and representative motion state features during the monorail crane's operation are extracted and labeled to directly identify specific events or operating modes occurring during operation. This provides a unified, structured input data with the same dimensions and meaning for subsequent state assessments of different travel distances and cranes.

[0081] Furthermore, the originally isolated instantaneous state vectors are linked together through a time axis to form a state evolution record with a clear temporal order and continuity. This provides a standardized and structured input format for all subsequent algorithms and models that require time series analysis, pattern recognition, and state prediction. At the same time, organizing data in matrix form is more efficient and compact, facilitating database storage, fast retrieval, and batch processing operations.

[0082] In summary, the equidistant trajectory sequence is a highly standardized data format that is easy to receive, store, process, and visualize by various backend systems, facilitating system integration and upper-level application development. It also provides high-fidelity, real-time position-driven data for building a digital twin of a monorail, enabling precise synchronization between the virtual and physical worlds. Furthermore, it can be combined with vibration, current, and other sensor data analysis to analyze the equipment's operating status at specific locations, providing spatiotemporal correlation information for predictive maintenance.

[0083] In summary, quantifying operational status and performance, identifying abnormal states such as excessively low peak acceleration, insufficient deceleration or excessively long braking distance, and high-frequency acceleration fluctuations, and calculating the deviation between the current state vector and the healthy baseline in real time provides early warnings for equipment maintenance. This, combined with information such as lifting point location and startup acceleration, helps infer the load magnitude. Furthermore, after a fault occurs, it can retrospectively analyze the state vector sequence prior to the fault to identify abnormal operating patterns or equipment performance degradation trends that led to the fault. Compared to storing complete, equally spaced trajectory sequences, storing state vectors significantly saves storage space and bandwidth. When analyzing large amounts of historical data, advanced analyses such as clustering, classification, and association rule mining can be performed directly at the state vector level, which is far more efficient than processing raw trajectory data and can quickly discover patterns.

[0084] In summary, by analyzing the row sequences of the state matrix, the typical stages of monorail operation can be automatically identified. Observing the distribution, frequency, and intensity changes of acceleration and deceleration envelope peaks on the time axis allows for quantitative analysis of operational patterns and stability across different road sections and time periods. It also supports advanced anomaly detection and diagnosis, not only detecting outliers at individual time points but also anomaly patterns. This enables more accurate prediction of the monorail's future task completion accuracy and optimizes scheduling plans.

[0085] In this embodiment of the invention, when the abnormal feature vector generation module performs fault feature extraction on the operating state matrix to obtain the abnormal feature vector of the monorail crane, it is specifically used for:

[0086] Discrete wavelet decomposition is performed on the operating state matrix to obtain the wavelet coefficient matrix of the monorail crane;

[0087] Singular value decomposition is performed on the wavelet coefficient matrix to obtain the amplitude domain feature vector of the monorail crane;

[0088] The variance threshold filtering of the amplitude feature vector is used to obtain the abnormal feature vector of the monorail crane.

[0089] Specifically, the original signal is captured by low-pass filtering and downsampling to obtain the low-frequency trend and main outline of the signal, and then high-pass filtering and downsampling are used to capture the high-frequency details, transient changes and noise of the signal. At the same time, the approximation coefficient sequence is recursively decomposed to the next level, thereby obtaining new approximation coefficients at lower frequencies and new detail coefficients at higher frequencies. Finally, after decomposing each feature of the state matrix, the approximation coefficients and detail coefficients obtained from each level are reorganized according to the number of decomposition levels and types to finally form a wavelet coefficient matrix.

[0090] Specifically, based on the obtained wavelet coefficient matrix of the monorail crane, the high-dimensional wavelet coefficient space that may have multiple correlations is reduced in dimension and features are extracted. The intrinsic structure of the data is revealed by SVD, and a set of mutually orthogonal basis vectors, representing the most essential linear combination pattern between the original features and wavelet coefficients, and sorted by their singular values, are output as amplitude domain feature vectors.

[0091] Specifically, a large number of amplitude feature vector samples known to be in normal states are collected. The variance of each amplitude feature vector component in these normal samples is calculated to represent the normal range of natural fluctuation of that component in a healthy state. A variance threshold is set for each component. This threshold is typically determined based on the variance distribution of normal samples. When processing a new amplitude feature vector corresponding to a new state to be detected, the squared deviation of each component of the new vector from the mean of the healthy baseline is calculated. The squared deviation is compared with the variance threshold of the corresponding component, and all components that meet the threshold are selected.

[0092] Furthermore, it distinguishes slowly changing background trends from instantaneous events and high-frequency noise, allowing subsequent analysis to operate on specific frequency ranges, improving the targeting and anti-interference capabilities of the analysis. At the same time, it transforms the original time-domain signal into the wavelet domain. The wavelet coefficients themselves contain all the information of the original signal, but their sparsity and energy concentration characteristics make information representation more efficient, facilitating compression and key information extraction. It has both time-domain and frequency-domain localization capabilities, laying the foundation for understanding when and at what frequency the state characteristics change.

[0093] Furthermore, the high-dimensional and strongly correlated wavelet coefficient space is compressed into a set of low-dimensional core feature vectors that contain the majority of information, automatically discovering and quantifying the hidden linear dependencies and combination patterns among the original numerous features—wavelet coefficients. This generates mutually orthogonal amplitude-domain feature vectors, simplifying subsequent statistical analysis, regression, and classification models.

[0094] Furthermore, the amplitude feature vector representing the overall status of the equipment is finely filtered to remove noise components that belong to normal background fluctuations, retaining only signal components that strongly indicate potential anomalies. This significantly improves the signal-to-noise ratio of subsequent anomaly detection or diagnosis, making it easier for subsequent algorithms to identify genuine fault signals. The feature dimension is further reduced, retaining only the most relevant and suspicious components, simplifying subsequent calculations and providing a unified format of feature vector input containing only anomaly information for subsequent modules such as anomaly scoring, fault classification, and alarm triggering.

[0095] In general, different types of equipment faults produce significant anomalous energy at specific wavelet decomposition layers. Analyzing the detail coefficients of corresponding features in the wavelet coefficient matrix at specific layers can accurately locate the characteristic frequencies of the fault, greatly improving the accuracy and specificity of fault type diagnosis. Meanwhile, the initial impact signals generated in the early stages of a fault are weak and often drowned out by strong background noise. Multi-scale analysis can amplify the detail coefficients of specific frequency bands, enabling the effective extraction and identification of weak impact signals with characteristic frequencies, thus achieving early warning.

[0096] In summary, comparing the values ​​of the main amplitude domain feature vectors of a monorail crane across different time periods and their derived low-dimensional features can efficiently and robustly identify the equipment's operating status category. Simultaneously, it automatically identifies feature combinations sensitive to faults. If a certain fault specifically alters the correlation of a set of features—wavelet coefficients—then the change in the corresponding amplitude domain feature vector or component becomes a strong indicator for diagnosing the fault. This is because different fault types may trigger different hidden patterns. Analyzing which amplitude domain feature vectors change significantly when a fault occurs helps distinguish the fault type.

[0097] In summary, setting an anomaly score threshold enables real-time automated anomaly alarms. Variance filtering retains only the components that truly reflect abnormal fluctuations, reducing the risk of false alarms. Furthermore, each component in the anomaly feature vector corresponds to a specific SVD combination pattern. Combining this with the original features—wavelet coefficients—that have high component weights allows for further tracing of the original physical parameters and frequency bands leading to the anomaly of that pattern, providing a foundation for root cause analysis. Since the variance threshold is dynamically calculated based on health status data, it automatically adapts to individual equipment differences and normal fluctuation levels at different operating stages (such as reduced fluctuations after the break-in period).

[0098] In this embodiment of the invention, when the abnormal feature vector generation module performs waveform characteristic quantization on the wavelet coefficient matrix to obtain the amplitude domain feature vector of the monorail crane, it is specifically used for:

[0099] The wavelet coefficient matrix is ​​dimensionally segmented to obtain the independent frequency band vibration matrix of the monorail crane;

[0100] The amplitude modulus of the independent frequency band vibration matrix is ​​taken to obtain the amplitude domain statistics of the monorail crane;

[0101] Based on the spectral kurtosis, the amplitude domain statistics are demodulated using the resonance band to obtain the amplitude domain feature sequence of the monorail crane;

[0102] Tensor folding is performed on the amplitude-domain feature sequence to obtain the amplitude-domain feature vector of the monorail crane.

[0103] Specifically, based on the wavelet decomposition level and sampling frequency, the center frequency or frequency range corresponding to the detail coefficients of each level and the highest level approximation coefficients is determined. For each selected target frequency band, all vibration coefficient columns belonging to that frequency band are extracted. These extracted vibration coefficient columns belonging to the same frequency band are then combined into a new matrix according to their original time order. The number of rows (time points) in this new matrix is ​​the same as the wavelet coefficient matrix, and the number of columns equals the number of vibration coefficient columns extracted for that frequency band. Based on the physical meaning of frequency, the vibration information contained in the high-dimensional wavelet coefficient matrix is ​​decoupled and recombined by frequency band, providing a structured data foundation for subsequent refined vibration analysis focused on specific frequency ranges.

[0104] Specifically, after performing the above amplitude calculation on all elements, a new matrix with the same dimensions as the original matrix is ​​obtained. The elements of the matrix represent the instantaneous vibration energy of the vibration component at the corresponding time point in the frequency band. Then, the amplitude information of the wavelet coefficients representing the time-varying energy of the vibration component in a specific frequency band is compressed and condensed into a set of scalar statistical indicators that can characterize the vibration intensity, distribution characteristics, and impact characteristics of the frequency band and have clear physical meaning and diagnostic value.

[0105] Specifically, the frequency domain kurtosis sensitivity of the spectral kurtosis is used to automatically and intelligently locate the resonant frequency band with the highest signal-to-noise ratio that is most sensitive to the impact in the original vibration signal. The optimal resonant frequency band signal is demodulated with Hilbert envelope, the high-frequency resonant carrier is stripped off, and the low-frequency envelope signal containing fault characteristic frequency information is extracted. The amplitude domain statistics of the envelope signal are calculated, and the demodulated impact modulation information is quantized into a set of scalar features with clear physical meaning and diagnostic value. Finally, the above process is repeated in a continuous time window to generate a feature sequence that reflects the change of impact characteristics over time.

[0106] Specifically, by introducing new analytical perspectives such as the structural dimension that characterizes the inherent spatial layout of the equipment, the positional dimension that reflects the specific coordinates of the equipment in the track network, and the operating mode dimension that represents the different working states of the equipment, the serialized data that could only describe the amplitude-frequency dynamic evolution characteristics of the equipment state over time and position is deeply reorganized and constructed into a high-dimensional tensor data structure containing complex correlation information between the equipment in multi-dimensional spatiotemporal and operating modes. This enables a more condensed representation of the spatiotemporal evolution law of the equipment state, and this high-dimensional tensor can be further vectorized to provide structured input features for downstream machine learning modeling and intelligent analysis tasks.

[0107] Furthermore, it overcomes the limitation of wavelet coefficient matrices mixing all frequency band information together, realizes the physical separation of vibration signals in the frequency domain, and allows subsequent analysis to independently and deeply study the vibration behavior in a specific frequency band. The original mixed coefficient matrix is ​​organized into multiple smaller and more focused matrices according to physical meaning, i.e., frequency band, which facilitates parallel processing and targeted storage.

[0108] Furthermore, key diagnostic features that are widely proven to be sensitive to mechanical conditions are independently extracted from the frequency band vibration analysis pipeline. The high-dimensional, time-varying frequency band vibration amplitude sequence is compressed into a set of low-dimensional scalar features with strong characterization capabilities that summarize the overall behavioral characteristics of the amplitude sequence throughout the entire analysis period. This simplifies the subsequent state classifier, fault diagnosis model, and health assessment algorithm by providing a unified format, fixed dimension, and clearly defined physical meaning of feature vector input.

[0109] Furthermore, following the basic vibration signal processing, advanced diagnosis is performed for impact-type mechanical faults. By focusing on the optimal resonance band and envelope demodulation to suppress background noise and interference unrelated to the impact, the detectability of fault characteristic frequencies is improved. The demodulated envelope signal is transformed into standardized amplitude domain statistical features, providing direct input for subsequent fault diagnosis, severity assessment, and trend analysis, and providing structured data for analyzing the temporal evolution of impact characteristics.

[0110] Furthermore, by constructing a comprehensive standardized data representation framework that integrates the spatial topological heterogeneity of equipment, the dynamic characteristics of track position migration trajectories, and the response patterns under the constraints of operating mode switching, a unified representation system that carries the entire spatiotemporal context is provided to characterize the non-uniformity of equipment status in spatial distribution, the evolution path trend in continuous track displacement, and the specific response patterns under different operating conditions. This allows for the deep integration and systematic reconstruction of the amplitude-domain feature sequence data, which might have been recorded at discrete points or stored in fragments according to independent operating modes, based on logical dimensions such as multi-dimensional spatial structure, motion trajectory coordinates, and operating mode. Ultimately, this results in a highly structured standardized tensor organization with strict dimensional alignment specifications, improving the systematic management efficiency and cross-dimensional correlation computing capabilities of heterogeneous state data.

[0111] In summary, obtaining independent vibration matrices for the target frequency band through dimensional segmentation and eliminating interference from noise in other irrelevant frequency bands greatly improves the sensitivity and accuracy of detecting weak fault characteristic signals within that frequency band. Furthermore, vibration signals from different frequency bands contain different information, allowing for the extraction of features most suitable for each independent frequency band vibration matrix. Specifically, the high-frequency band is suitable for extracting impact indicators and envelope spectrum features, the mid-frequency band is suitable for extracting features related to gear meshing and structural resonance, and the low-frequency band is suitable for extracting features related to imbalance, asymmetry, foundation loosening, and operating speed.

[0112] In summary, by comparing the calculated amplitude domain statistics with preset health baseline thresholds or historical trends, it is possible to quickly and automatically determine whether the vibration in the frequency band is abnormal. The anomaly detection calculation efficiency based on amplitude domain statistics provides data support for subsequent real-time execution on embedded systems or edge devices. At the same time, different fault types will cause different amplitude domain statistics change patterns in specific frequency bands. By using these patterned statistical changes as feature inputs to the classifier, automated fault type identification can be achieved.

[0113] In summary, by using SK as an intelligent guide, the high signal-to-noise ratio resonant frequency band most sensitive to impact in vibration signals is automatically locked. Envelope demodulation technology is used to strip the high-frequency carrier, extract the low-frequency envelope signal containing fault characteristic frequencies, and calculate its amplitude domain statistics to form time-series characteristics. This provides a powerful core solution based on signal processing and statistics for early high-sensitivity detection of weak impact faults, accurate exposure of fault characteristic frequencies and diagnosis types, quantification of fault severity and evolution trends, improvement of diagnostic automation and reliability, and empowerment of intelligent prediction models, enabling predictive health management of key rotating components of monorail cranes.

[0114] In summary, by introducing key dimensions such as spatial location, track segment, or operating mode, the amplitude-domain feature sequence describing equipment state evolution is restructured into a structured representation containing rich spatiotemporal correlations and modal information. This provides a powerful and structured data foundation for comprehensively characterizing the spatial distribution of equipment states, revealing the evolution patterns along the track, distinguishing the impact of operating modes, improving the accuracy of fault location and diagnosis, enabling tensor decomposition and deep learning, constructing digital twin state interfaces, and supporting full lifecycle data analysis.

[0115] In this embodiment of the invention, when the drive unit vibration spectrum generation module performs adversarial spectrum reconstruction on the abnormal feature vector to obtain the drive unit vibration spectrum of the monorail crane, it is specifically used for:

[0116] The abnormal feature vector is subjected to windowed noise separation to obtain the background noise suppression signal of the monorail crane;

[0117] The phase-compensated correction is performed on the noise floor signal to obtain the phase-aligned spectrum of the monorail.

[0118] The phase-aligned spectrum is reconstructed by performing adversarial enhancement to obtain the vibration spectrum of the drive unit of the monorail crane;

[0119] Specifically, the high-dimensional feature vectors collected by discretization to characterize abnormal equipment states are regarded as non-stationary signals with time-series evolution characteristics. Through adaptive windowing time-frequency analysis technology, the multi-source random noise components superimposed on the original features within each local time window are statistically estimated and blind source separated, thereby effectively filtering out the mixed random fluctuations introduced by environmental interference, measurement errors and system uncertainties. Finally, the purified feature signals with highly focused spatial-energy distribution on the potential deterministic anomaly generation mechanism and significant physical interpretability are extracted.

[0120] Specifically, by precisely identifying and dynamically correcting the composite phase distortion field formed by the coupling of multiple interference factors in the residual signal after background noise suppression, including the non-ideal frequency response characteristics of the sensing device, the group delay distortion introduced by the preprocessing stage, the path-related phase shift caused by multipath propagation effects, and the rotation period synchronization error caused by insufficient accuracy of the speed sensor, a spectrum representation with strict physical consistency is finally reconstructed. This spectrum can not only more realistically map the dynamic characteristics of the internal excitation mechanism of the signal source in the joint distribution of amplitude and phase in the frequency domain, but also achieve high-precision synchronization alignment with the reference signal in the time and frequency domain.

[0121] Specifically, multiple vibration signals or different time segments of the same signal, which may have experienced phase shifts due to sensor location, trigger time, or signal propagation delay, are subjected to phase calibration using specific signal processing algorithms, namely cross-correlation analysis, phase synchronization, and phase-locked loop (PLL) technology. The result is a spectrum in which the phase information of different signals or different time segments is consistent and comparable, eliminating interference caused by phase differences and making the spectrum more reflective of the true periodic characteristics of the signal.

[0122] Furthermore, random fluctuation noise that may exist in the abnormal feature vector caused by non-fault factors is suppressed to improve the signal-to-noise ratio of subsequent analysis. Random interference that may not have been completely eliminated in variance threshold filtering is further purified to highlight the true abnormal patterns. This makes the patterns representing potential faults or abnormal states continuous, stable and identifiable in the signal, reducing misjudgments caused by noise and providing cleaner and more reliable input signals for subsequent fault mode recognition, classifiers, trend analysis and other modules.

[0123] Furthermore, it suppresses the impact of phase distortion on the accuracy of spectrum analysis, ensures the physical accuracy and interpretability of the spectrum results, recovers valuable phase information in the signal, and provides a reliable data foundation for diagnostic methods that rely on phase relationships. Through order tracking and angle domain resampling, it solves the spectral leakage and phase ambiguity caused by speed fluctuations, and provides phase-correct input spectra for all subsequent diagnostic algorithms that require accurate spectral information.

[0124] Furthermore, the system receives signals that have undergone preliminary processing and key corrections, and outputs a spectrum highly optimized by adversarial mechanisms. This provides more reliable and sensitive basic data for subsequent fault feature identification, status assessment, diagnostic algorithms, or health indicator calculations, while also improving data quality and empowering subsequent diagnostic modules to more easily detect weak fault features and reduce false alarms and missed alarms.

[0125] In summary, by utilizing local window analysis techniques, we can estimate and remove stray random background noise from discrete anomalous feature signals, thereby improving the signal-to-noise ratio and clarity of anomalous patterns. Simultaneously, we provide purified anomalous signal inputs to enhance fault diagnosis accuracy and reliability, strengthen the detection capability of weak faults, improve feature interpretability, increase trend analysis accuracy, optimize classifier performance, achieve adaptive noise reduction, and support more accurate fault severity assessment.

[0126] In summary, this method corrects phase distortion introduced during signal acquisition, processing, and propagation, restores the true phase relationship between signal components, or achieves synchronization with a reference signal, generating physically more accurate and reliable spectrum diagrams. Simultaneously, it provides a foundation of phase-correct spectral data for precise sideband phase analysis, improved coherent path analysis accuracy, accurate mode recognition, enhanced acoustic diagnostics, elimination of the effects of rotational speed fluctuations, optimization of multi-point joint analysis, and empowerment of advanced intelligent diagnostic algorithms.

[0127] In summary, by combining phase-aligned signal processing techniques with the intelligent enhancement capabilities of generative adversarial networks, the original vibration signal is transformed into a high-quality spectrum with lower noise, more prominent features, consistent phase, and a high degree of realism reflecting the vibration state of the monorail crane's drive unit. This ensures the accuracy and reliability of subsequent analysis, enabling precise and robust equipment condition monitoring and maintenance decisions.

[0128] In this embodiment of the invention, when the diagnostic element generation module performs wavelet packet entropy decomposition on the vibration spectrum of the drive unit based on the load weight signal and real-time battery status signal of the monorail to obtain the diagnostic elements of the monorail, it is specifically used for:

[0129] Harmonic suppression is applied to the vibration signal of the drive unit to generate a baseline stable signal.

[0130] Based on the baseline stable signal, the load weight signal and the battery status signal are synchronously aligned, and vibration data segments are output.

[0131] The vibration data segment is divided into frequency bands based on orthogonal wavelet basis to obtain the time-frequency sub-band signal of the monorail.

[0132] The time-frequency sub-band signal is recursively quantized based on phase space reconstruction to obtain the frequency band entropy value of the monorail.

[0133] The frequency band entropy values ​​are sorted in ascending order based on the frequency band number to obtain the complexity distribution feature vector of the monorail. The complexity distribution feature vector is then jointly encoded to obtain the diagnostic primitive of the monorail.

[0134] Specifically, the spectral peaks in the signal corresponding to the main rotating components of the drive unit are identified. Specific signal processing is applied to selectively attenuate or filter out the energy of the identified fundamental frequency and harmonic frequency components. The filter parameters are automatically adjusted at the harmonic frequency points to deeply suppress the energy near those frequencies. The magnitude and phase of the harmonic components are directly estimated in the frequency domain and then subtracted from the original spectrum. When the signal is highly periodic, precise synchronous averaging can effectively retain components that are strictly synchronized with the reference axis and average out asynchronous components.

[0135] Specifically, using the baseline stable signal as a common and reliable time reference axis, the load weight signal and battery status signal are adjusted on the time axis so that their key event points are precisely matched with the corresponding feature points of the baseline stable signal. After the above alignment is completed, vibration data segments that completely correspond to the aligned load and battery status time periods are extracted.

[0136] Specifically, the input vibration data segment is decomposed into multiple levels. The signal is filtered and downsampled using a pair of orthogonal wavelet filters. At each level of decomposition, the approximation coefficients and detail coefficients are decomposed simultaneously. Specific frequency ranges are located more accurately through high-resolution partitioning. The decomposed sub-band signal corresponds to the components of the original signal within a specific frequency range, preserving the time-frequency localization capability of these frequency components as they change over time. The final output is the reconstructed sub-band signal.

[0137] Specifically, based on the differential homeomorphism embedding theorem, the one-dimensional sub-band signals obtained by time-frequency decomposition are mapped to a high-dimensional phase space through the delay coordinate reconstruction method. Dynamic orbits with topological invariance are constructed according to the embedding dimension and time delay parameters. Recursive quantitative analysis is used to perform neighborhood density statistics on the topological structure of the multidimensional trajectory in the reconstructed phase space to quantify its recursive graph texture features. Finally, a set of nonlinear dynamic indices distributed across frequency bands is generated.

[0138] Specifically, the entropy values ​​of the monorail vibration signal calculated in each sub-band are monotonically rearranged in ascending order according to the frequency band number to generate a feature vector representing the gradient distribution of frequency band complexity. Through coding techniques such as symbolic aggregation approximation and convolutional dictionary learning, the numerical distribution pattern of this vector is nonlinearly reduced in dimension and structured compressed, and finally outputs a discrete diagnostic primitive symbol sequence with strong class separability.

[0139] Furthermore, by suppressing strong harmonic background, the abnormal vibration characteristics with small amplitude and frequency independent of harmonics are highlighted, thus amplifying the signal-to-noise ratio of potential faults. This provides more accurate signal input for subsequent fault feature extraction algorithms, focusing on detecting and quantifying non-stationary, non-periodic, and non-harmonic characteristics that truly reflect potential damage or anomalies.

[0140] Furthermore, precise time alignment is a prerequisite for this type of multi-dimensional data fusion and correlation analysis, enabling the identification of the load source, battery state, and time misalignment caused by a particular vibration feature. Simultaneously, the aligned, high-quality samples serve as reliable input data for all subsequent analyses, ensuring that the analysis results reflect the true load-battery vibration relationship.

[0141] Furthermore, wavelet basis functions with tight support characteristics are employed to perform multi-resolution orthogonal decomposition of the original mechanical vibration signal covering a wide frequency band in the frequency-scale domain, accurately separating it to the critical sub-band with physical significance. Through wavelet transform, a dynamically adjustable time-frequency window structure is used on the time-frequency plane. The low-frequency band automatically widens the time window to capture the macroscopic fault features of continuous slow fluctuations, while the high-frequency band compresses the time window to focus on the microsecond-level details of transient impacts. This adaptively matches the long-period slow-varying characteristics of the low-frequency modulation component and the short-time abrupt change behavior of the high-frequency resonance / impact component in the rotating machinery vibration signal, establishing an optimal time-frequency observation basis for multi-scale fault feature extraction.

[0142] Furthermore, going beyond traditional time-domain and frequency-domain statistical features, we extract the system dynamic behavior features contained in the sub-band signals of each frequency band, focusing on quantifying the complexity, regularity, and predictability of the vibration dynamic state within each frequency band. This provides a set of features that are highly sensitive to changes in equipment state for subsequent diagnostic models, and transforms the physically meaningful time-frequency localized information obtained from wavelet decomposition into entropy values, which are quantitative indicators reflecting the dynamic health state within that frequency band.

[0143] Furthermore, by employing a collaborative dimensionality reduction strategy based on feature selection using the maximum correlation and minimum redundancy criterion and manifold learning, the original high-dimensional time-frequency entropy feature matrix is ​​compressed into a key low-dimensional ordered vector that preserves the complexity order relationship between frequency bands, eliminating information redundancy caused by environmental noise. Subsequently, using a joint encoding framework of symbolic aggregation approximation and dictionary learning, the continuous complexity distribution pattern is transformed into a discrete primitive symbol sequence with physical semantics, constructing a nonlinear mapping bridge from the vibration feature space to the fault mode space. Finally, input features that are both lightweight and highly interpretable are generated, providing a traceable decision basis for the identification of degradation states based on hidden Markov models.

[0144] In summary, by accurately identifying and suppressing the strong periodic harmonic components generated by normal rotation and meshing in the vibration signal of the drive unit, and stripping away strong background interference, a stable baseline signal is generated, amplifying the weak anharmonic and non-stationary features hidden in the background that are related to potential faults. This improves the sensitivity, accuracy, and reliability of subsequent feature extraction and diagnostic algorithms, laying a solid foundation for early detection of potential problems, accurate judgment of fault type and severity, and effective predictive maintenance.

[0145] In summary, by leveraging the superior time-frequency localization capability and redundancy-free frequency band segmentation characteristics of orthogonal wavelet bases, the vibration signal of a monorail crane drive unit, which contains a mixture of multiple components, is decomposed into a set of time-frequency sub-band signals focused on different frequency ranges. This approach accurately separates the fault characteristics of different components, significantly enhances the detectability of weak transient impacts, effectively separates concurrent faults, deeply analyzes modulation phenomena, adapts to non-stationary signals, and provides structured, high-value input for subsequent feature extraction and intelligent diagnostic algorithms. Ultimately, this achieves precise component-level fault diagnosis and condition assessment, improving the diagnostic capabilities and reliability of the entire monorail crane drive unit health management system.

[0146] In summary, the aligned data solved the time synchronization problem of multi-source heterogeneous sensor data, laying a solid and reliable data foundation for subsequent in-depth analysis of the intrinsic relationship between load, battery status and equipment vibration characteristics. This ensures that the analysis results reflect the true physical causal relationship, rather than the illusion caused by time misalignment, thereby greatly improving the effectiveness, accuracy and credibility of the entire solution.

[0147] In summary, this method provides highly sensitive indicators for detecting early, subtle faults, enabling the location of fault bands, differentiation of fault types, quantification of dynamic degradation levels, and effective characterization of transient and non-stationary signal properties. The complementary nature of band entropy and band energy, among other traditional features, provides key nonlinear dynamic characteristics for building more robust and powerful intelligent diagnostic models, thus contributing to improved early warning capabilities and diagnostic accuracy of monorail crane drive unit condition monitoring and predictive maintenance systems.

[0148] In summary, ascending order highlights the entropy gradient difference between the low-frequency and high-frequency bands, enhancing sensitivity to early wear and sudden damage. When the bearing raceway peels off, the entropy value of the high-frequency band rises sharply, causing a sudden change in the amplitude of the vector tail, which is different from the smooth distribution in the normal state. The joint encoding absorbs the influence of speed fluctuations, and a warning is triggered when a high-risk primitive combination is detected.

[0149] In this embodiment of the invention, when the diagnostic primitive generation module performs recursive quantization of the time-frequency subband signal based on phase space reconstruction to obtain the frequency band entropy value set of the monorail, it is specifically used for:

[0150] The time-frequency subband signal is encapsulated into a two-order vector to obtain the basic trajectory and extended trajectory of the monorail crane;

[0151] The basic trajectory and the extended trajectory are recursively scanned to obtain the dual similarity measure of the monorail crane;

[0152] Subband aggregation is performed on the dual similarity measure to obtain the frequency band entropy value set of the monorail crane.

[0153] Specifically, the original high-dimensional time-frequency subband signal is transformed into a structured low-dimensional vector representation that strictly satisfies the axiom of metric space through equidistant embedding mapping. While maintaining the topological invariance of the time-frequency energy distribution, sparse coding is used to decouple the coupled vibration modes of the mechanical system, so that the bearing damage characteristics, gear meshing characteristics and track resonance characteristics can be orthogonally separable in the vector space, completely eliminating the spectral aliasing effect and operating noise interference of multi-source excitation signals. Finally, a lightweight feature base for multi-task sharing such as degradation state assessment and fault mode classification is constructed.

[0154] Specifically, based on a hybrid measure of recursive quantitative analysis and dynamic time warping, the self-similarity coefficient of the operating mode, periodic track stability index, and time scale of the multi-source sensor signals of the monorail are quantified on the time axis. Through recursive comparison of phase space trajectories, the periodicity of gear meshing harmonics and the quasi-periodic reproducibility of the impact of the bearing wheelset on the track joint are accurately identified in the current fluctuations of the drive motor. Based on the dynamic similarity analysis of multidimensional time series, the chaotic synchronization characteristics between the pressure pulsation of the braking system and the sway of the basket are revealed, providing cross-scale correlation diagnostic basis for early degradation.

[0155] Specifically, the dual similarity measures within the time-frequency sub-band are aggregated into a sub-band comprehensive similarity scalar through weighted fusion (such as the entropy weight method); based on entropy theory, the dynamic complexity entropy value of the sub-band comprehensive similarity sequence within the sliding time window is calculated, and finally, a set of entropy values ​​distributed across frequency bands is output to quantify the temporal unpredictability and mode disorder of vibration signals in each frequency band.

[0156] Furthermore, as the core data engine of the digital twin system for the entire life cycle of a monorail crane, this structured feature vector constructs a unified feature representation framework covering the drive system, load-bearing structure, and running track by integrating multi-scale dynamic fingerprints in the time and frequency domains. This provides real-time decoupled observation capabilities of bearing damage index and track smoothness spectrum for condition monitoring; forms a symbolic primitive topology that can map to the fault tree ontology at the fault diagnosis level; supports a high-precision Kalman filter state-space model for trajectory prediction; and ultimately provides the gradient descent direction of the minimum energy control law for the control optimization system, determining the industrial-grade intelligent closed-loop foundation from perception to decision-making.

[0157] Furthermore, by quantifying the topological consistency between the actual operating trajectory and the benchmark template in phase space, the driving stability, track following accuracy, and load swing suppression performance are accurately evaluated. High similarity indicates that the system is in a controlled steady state; low similarity reveals potential anomalies. These two measures, refined through nonlinear dimensionality reduction, carry the core dynamic characteristics of the system's Lyapunov characteristic exponent change rate and attractor fractal dimension, and can be directly input into the SVM state classifier to form a closed-loop intelligent operation and maintenance chain from perception to decision based on the fault diagnosis tree.

[0158] Furthermore, the high-dimensional bisimilarity matrix is ​​compressed into a low-dimensional entropy vector, preserving the frequency band-specific dynamic essence. By suppressing environmental noise through sub-band aggregation, the feature signal-to-noise ratio is improved, and the time-domain waveform of microscopic signals is linked to the frequency-domain complexity of macroscopic systems, providing a unified analytical framework for high-frequency entropy increase and low-frequency entropy decrease.

[0159] In summary, by performing two-order vector encapsulation on the time-frequency subband signal to obtain the basic trajectory and extended trajectory, the monorail operation information contained in the original signal is extracted in a structured and hierarchical manner, completing the transformation from complex data to usable features. This provides a clear and hierarchical data foundation for subsequent state analysis, fault diagnosis, and intelligent control algorithms; it improves the availability of information, the depth of analysis, and the processing efficiency, supporting the reliable and intelligent operation of the monorail.

[0160] In summary, the quantification of similarity from static features to dynamic patterns has been completed; direct criteria have been provided for assessing core operational stability and diagnosing mechanical health status and early anomalies; essential insights into the dynamic characteristics of the system have been achieved, key technical support has been provided for early and accurate fault detection and predictive maintenance, and the safety and operational efficiency of monorail cranes have been improved.

[0161] In summary, by capturing the dynamic instability of micro-impact signals caused by bearing spalling, the similarity of time-domain waveforms and the stability of phase-space behavior are transformed into a frequency band complexity spectrum, enabling the equipment degradation information implicit in the vibration signal to be explicitly expressed in the form of entropy gradient, providing a core feature base with both physical interpretability for industrial diagnostics.

[0162] In this embodiment of the invention, when the traction force dynamic allocation scheme and multi-unit self-healing strategy generation module performs fault unit decoupling on the diagnostic primitives based on the fault rule base to obtain the traction force dynamic allocation scheme and multi-unit self-healing strategy of the monorail, it is specifically used for:

[0163] The diagnostic primitives are subjected to feature dimensionality reduction to obtain the set of independent fault units of the monorail.

[0164] The energy flow path is reconstructed for the set of independent fault units to obtain a dynamic traction force distribution view;

[0165] Multimodal fusion is performed on the dynamic traction force distribution view to obtain the adaptive distribution matrix of the monorail.

[0166] By redirecting the bus signal of the adaptive allocation matrix, the multi-link self-healing strategy of the monorail crane is obtained;

[0167] Specifically, numerous statistical features, time-domain features, and frequency-domain features extracted from sensor signals such as vibration, sound, current, and temperature are mapped to a low-dimensional space. Highly correlated and redundant features, as well as features that are mainly noise, are identified and eliminated. The core information that best represents the intrinsic structure of the original data and can most effectively distinguish different fault modes is retained or constructed.

[0168] Specifically, discrete independent fault units are abstracted into directed weighted nodes. Based on the physical connection relationship of the transmission chain and real-time power sensor data, the energy transfer path under fault conditions is reconstructed. The traction force contribution weight of each node is calculated through a dynamic load distribution algorithm, and a thermal layer overlay view that integrates spatial location, fault influence coefficient and real-time traction ratio is generated to present the system-level traction redistribution process caused by local failure.

[0169] Specifically, physical state observations of different modes are mapped to a unified mathematical representation, and the optimal traction force weight coefficient of each execution unit is dynamically calculated through deep learning or optimization algorithms to form matrix control commands.

[0170] Specifically, when the system detects an actuator-level fault or a communication link fault, it identifies the failure point, recalculates the traction command allocated to the failed unit in the adaptive allocation matrix, and maps it to a preset, healthy backup actuator or alternative communication path. It dynamically modifies the routing table of the communication network and uses hardware redundancy channels to physically or logically send the control signals originally sent to the failed unit to a healthy unit that can replace its function, ensuring that the control command can always reach the node that needs to be executed, even if the original path is interrupted.

[0171] Furthermore, the massive features extracted from the original signal, which may contain a lot of irrelevant or redundant information, are transformed into refined fault indicators with high information density, laying the foundation for subsequent fault diagnosis, classification, and prediction models.

[0172] Furthermore, component-level faults are mapped to drive train energy flow losses, the impact of faults on system-level traction capability is quantified, and traction force redistribution optimization boundaries are provided for the control system. This links mechanical fault diagnosis with electric drive control, enabling a closed loop of damage identification, energy flow reconstruction, and control response.

[0173] Furthermore, by connecting the fault diagnosis domain, energy scheduling domain, and control execution domain, when a unit fails, its load is migrated to a healthy unit in real time through matrix coefficient adjustment, providing a prototype of an iteratively optimizable control strategy for the digital twin.

[0174] Furthermore, upon detecting an execution-level fault, the dynamic reconfiguration of the control closed loop ensures that the chain of perception, decision-making, and execution is not interrupted by physical faults, enabling system failure operation. It intelligently calls upon potentially redundant actuators, backup communication channels, and other resources in the system, minimizing the impact of faults on system performance and operator perception through rapid and automated signal redirection and resource scheduling, thereby enhancing system transparency and trustworthiness.

[0175] In summary, generating a set of low-dimensional, uncorrelated core indicators that are strongly correlated with specific fault modes solves the computation, modeling, and interpretability problems caused by high-dimensional features, improves the efficiency, accuracy, robustness, interpretability, and fault isolation capabilities of the diagnostic system, and transforms complex equipment states into a concise, clear, and directly directional fault language that is easy to understand, monitor, and make decisions, thus realizing intelligent predictive maintenance.

[0176] In summary, transforming discrete fault events into a continuous energy rebalancing process ensures that the system can maintain the availability of its core output traction to the maximum extent even in the event of local component failure. At the same time, this process provides a foundation for the accurate construction of a dynamic mapping of the physical system state in digital twin systems, which reflects the cross-domain synergy of fault phenomena, energy flow dynamics, and control strategies.

[0177] In summary, multimodal perception enables accurate fault detection, adaptive matrix is ​​used to achieve coordinated rebalancing of energy and control, ultimately enabling the system to maintain steady-state traction force and providing a dynamic control gene library for closed-loop optimization for digital twins.

[0178] In summary, even when faced with underlying execution and communication failures, intelligent control decisions can still be effectively implemented through dynamic resource scheduling and path reconstruction, thereby enhancing the operational resilience and survivability of the monorail, expanding the system's fault tolerance, and achieving automatic fault isolation and functional compensation at the physical level. This ensures that the monorail can continue to operate safely and continuously even when local hardware and communication fail.

[0179] In this embodiment of the invention, when the joint control signal generation module performs instruction fusion of the dynamic traction force allocation scheme and the multi-link self-healing strategy to obtain the joint control signal of the monorail main control core, and wirelessly transmits the joint control signal to the monorail main control core, it is specifically used for:

[0180] The traction dynamic distribution scheme and the multi-link self-healing strategy are timestamped to obtain a synchronous command stream.

[0181] The control field is filled into the synchronization command stream to obtain the joint control signal frame of the monorail crane;

[0182] The control signal frame of the monorail is modulated and transmitted to obtain the radio frequency signal of the monorail.

[0183] The radio frequency signal is spatially guided and propagated to obtain the main control core execution command of the monorail crane.

[0184] Specifically, the allocation scheme generation module and the self-healing strategy generation module are ensured to use the same high-precision time source. The target effective timestamp of each traction force command is extracted from the dynamic allocation scheme. The traction force allocation command, signal redirection command, unit activation or disabling command, and parameter adjustment command are sorted according to the order of their target timestamps to identify whether there are logical errors or safety hazards caused by overlapping timestamps or reversed order.

[0185] Specifically, abstract synchronization command information is transformed into specific control codes recognizable by the device. Based on a predefined communication protocol and data dictionary, the synchronization command stream that needs to be sent is identified. According to the command type and parameters, the corresponding base value is searched or calculated. For each control element in the command stream, its fixed field position and corresponding code value specified in the protocol are found, and the calculated or found code value is accurately written into the specific control field position reserved in the joint control signal frame.

[0186] Specifically, the digital signal is converted into an analog signal suitable for wireless transmission, and the baseband signal spectrum is shifted to a designated radio frequency band. The modulated radio frequency signal is amplified to a sufficient power level to meet the requirements of communication distance and overcome environmental attenuation, and the electrical signal is converted into electromagnetic waves for transmission towards the target monorail equipment.

[0187] Specifically, by using an antenna with specific directionality, the radio frequency signal energy carrying the joint control signal frame is concentrated and projected toward the transmitting end where the target monorail is located. The signal is selectively received from the direction of the target monorail, which optimizes the signal propagation path in space and improves the signal-to-noise ratio of the receiver signal.

[0188] Furthermore, it achieves seamless coordination between dynamic traction control and fault self-healing reconfiguration. During the fault occurrence, activation of the self-healing strategy, and system state reconfiguration, it reduces performance fluctuations or momentary loss of control during state switching by precisely synchronizing control commands and reconfiguration commands, thus achieving smooth switching. It provides the underlying actuators and communication network with a command sequence that is strictly executed according to a timetable, giving the entire fault detection, decision-making, reconfiguration, and recovery control process a definite timeline, which facilitates performance evaluation and system verification.

[0189] Furthermore, the logical control commands issued by the upper-level dispatcher are transformed into low-level communication signals that the lower-level monorail controller can directly receive, parse, and execute. This provides standardized input for subsequent signal sending, transmission, reception, and decoding, and lays the foundation for the physical layer to send data and for the receiving end to perform frame synchronization, parsing, and verification.

[0190] Furthermore, it enables a wireless transmission channel for control information from the fixed end to the mobile end, transforming logical communication protocols into electromagnetic waves that can propagate in the physical world. This provides the foundation for the "wireless" capability of the entire control system, ensuring that the signal can effectively reach the range of the monorail and penetrate complex industrial environments.

[0191] Furthermore, by increasing the effective radiated power and receiving sensitivity in the target direction, space-guided control commands can reliably reach the maximum distance of the monorail. Simultaneously, for moving monorails, space-guided technology can dynamically adjust the beam direction to attempt continuous target tracking and maintain communication link continuity. This reduces the difficulty and range of signal interception by non-target receivers, providing a certain degree of physical layer confidentiality.

[0192] In summary, by eliminating the risk of timing disorder, ensuring causal dependence, and achieving deterministic execution, the ultimate timing barrier is built for monorail cranes to maintain stable, reliable, and safe operation under fault conditions. This is the core of the safe implementation of intelligent fault-tolerant control for complex electromechanical systems, ensuring that the system maintains deterministic behavior even in highly dynamic and complex fault recovery scenarios.

[0193] In summary, standardized encoding padding ensures that the sender and receiver have a completely consistent understanding of the meaning of each bit and byte in the control field, avoiding semantic ambiguity in the transmission of instructions. It maximizes the use of limited communication bandwidth, reduces redundant information, and improves the efficiency of instruction transmission, enabling high-level instructions issued by the upper-level system to be automatically and accurately translated into low-level machine language executable by the equipment, thus achieving automated integrated control of the monorail crane.

[0194] In summary, by converting specific radio technologies into radio frequency signals that can propagate in space, the system is provided with critical wireless connectivity, mobility support, communication reliability, and physical layer assurance, enabling all digital signal frames to be actually delivered to the target device.

[0195] In summary, by utilizing directional antenna technology to optimize the propagation of radio frequency signals in space toward the target monorail, and through energy focusing and interference suppression, the reliability, stability, communication distance, and anti-interference capability of the wireless control link are improved, ensuring that control commands can effectively reach the moving monorail equipment in complex industrial environments.

[0196] Reference Figure 2 The diagram shown is a flowchart illustrating a modularly integrated control method for a mine monorail crane according to an embodiment of the present invention. In this embodiment, the modularly integrated control method for a mine monorail crane includes:

[0197] S1. Spatiotemporal fusion of multi-source heterogeneous data of the monorail crane is performed to obtain the operating state matrix of the monorail crane;

[0198] S2. Extract fault features from the operating state matrix to obtain the abnormal feature vector of the monorail crane;

[0199] S3. Perform adversarial spectrum reconstruction on the abnormal feature vector to obtain the vibration spectrum of the drive unit of the monorail crane;

[0200] S4. Based on the load weight signal and real-time battery status signal of the monorail, perform wavelet packet sample entropy decomposition on the vibration spectrum of the drive unit to obtain the diagnostic primitives of the monorail.

[0201] S5. Based on the fault rule base, the diagnostic primitives are decoupled from the fault units to obtain the dynamic traction force distribution scheme and multi-link self-healing strategy of the monorail.

[0202] S6. The dynamic traction force distribution scheme and the multi-link self-healing strategy are fused together to obtain the joint control signal of the monorail main control core, and the joint control signal is wirelessly transmitted to the monorail main control core.

[0203] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0204] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0205] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A modularly integrated control system for a mining monorail crane, characterized in that, The system includes a running state matrix generation module, an abnormal feature vector generation module, a drive unit vibration spectrum generation module, a diagnostic element generation module, a traction force dynamic distribution scheme and multi-link self-healing strategy generation module, and a joint control signal generation module, among which: The operation status matrix generation module performs spatiotemporal fusion on the multi-source heterogeneous data of the monorail crane to obtain the operation status matrix of the monorail crane. The abnormal feature vector generation module extracts fault features from the operating state matrix to obtain the abnormal feature vector of the monorail crane. The drive unit vibration spectrum generation module performs adversarial spectrum reconstruction on the abnormal feature vector to obtain the drive unit vibration spectrum of the monorail crane. The diagnostic element generation module performs wavelet packet sample entropy decomposition on the vibration spectrum of the drive unit based on the load weight signal and real-time battery status signal of the monorail, and obtains the diagnostic elements of the monorail. The traction force dynamic distribution scheme and multi-link self-healing strategy generation module decouples the diagnostic primitives based on the fault rule base to obtain the traction force dynamic distribution scheme and multi-link self-healing strategy of the monorail. The joint control signal generation module fuses the dynamic traction force allocation scheme with the multi-link self-healing strategy to obtain the joint control signal of the monorail main control core, and wirelessly transmits the joint control signal to the monorail main control core.

2. The modularly integrated mine monorail control system as described in claim 1, characterized in that, When the operation status matrix generation module performs spatiotemporal fusion on the multi-source heterogeneous data of the monorail to obtain the operation status matrix of the monorail, it is specifically used for: Trajectory interpolation is performed on the multi-source heterogeneous data of the monorail to obtain the equally spaced trajectory sequence of the monorail; Envelope extraction is performed on the equally spaced trajectory sequence to obtain the state vector of the monorail crane; The state vectors are stacked in a time sequence to obtain the operating state matrix of the monorail crane.

3. The modularly integrated mine monorail control system as described in claim 1, characterized in that, When the abnormal feature vector generation module performs fault feature extraction on the operating state matrix to obtain the abnormal feature vector of the monorail crane, it is specifically used for: Discrete wavelet decomposition is performed on the operating state matrix to obtain the wavelet coefficient matrix of the monorail crane; Singular value decomposition is performed on the wavelet coefficient matrix to obtain the amplitude domain feature vector of the monorail crane; The variance threshold filtering of the amplitude feature vector is used to obtain the abnormal feature vector of the monorail crane.

4. The modularly integrated mine monorail control system as described in claim 3, characterized in that, When the abnormal feature vector generation module performs waveform characteristic quantization on the wavelet coefficient matrix to obtain the amplitude domain feature vector of the monorail crane, it is specifically used for: The wavelet coefficient matrix is ​​dimensionally segmented to obtain the independent frequency band vibration matrix of the monorail crane; The amplitude modulus of the independent frequency band vibration matrix is ​​taken to obtain the amplitude domain statistics of the monorail crane; Based on the spectral kurtosis, the amplitude domain statistics are demodulated using the resonance band to obtain the amplitude domain feature sequence of the monorail crane; Tensor folding is performed on the amplitude-domain feature sequence to obtain the amplitude-domain feature vector of the monorail crane.

5. The modularly integrated mine monorail control system as described in claim 1, characterized in that, When the drive unit vibration spectrum generation module performs adversarial spectrum reconstruction on the abnormal feature vector to obtain the drive unit vibration spectrum of the monorail crane, it is specifically used for: The abnormal feature vector is subjected to windowed noise separation to obtain the background noise suppression signal of the monorail crane; The phase-compensated correction is performed on the noise floor signal to obtain the phase-aligned spectrum of the monorail. The vibration spectrum of the drive unit of the monorail is obtained by performing adversarial enhancement reconstruction on the phase-aligned spectrum.

6. The modularly integrated mine monorail control system as described in claim 1, characterized in that, When the diagnostic element generation module performs wavelet packet entropy decomposition on the vibration spectrum of the drive unit based on the load weight signal and real-time battery status signal of the monorail to obtain the diagnostic elements of the monorail, it is specifically used for: Harmonic suppression is applied to the vibration signal of the drive unit to generate a baseline stable signal. Based on the baseline stable signal, the load weight signal and the battery status signal are synchronously aligned to obtain the vibration data segment; The vibration data segment is divided into frequency bands based on orthogonal wavelet basis to obtain the time-frequency sub-band signal of the monorail. The time-frequency sub-band signal is recursively quantized based on phase space reconstruction to obtain the frequency band entropy value of the monorail. The frequency band entropy values ​​are sorted in ascending order based on the frequency band number to obtain the complexity distribution feature vector of the monorail. The complexity distribution feature vector is then jointly encoded to obtain the diagnostic primitive of the monorail.

7. The modularly integrated mine monorail control system as described in claim 6, characterized in that, When the diagnostic primitive generation module performs recursive quantization of the time-frequency subband signal based on phase space reconstruction to obtain the frequency band entropy value set of the monorail, it is specifically used for: The time-frequency subband signal is encapsulated into a two-order vector to obtain the basic trajectory and extended trajectory of the monorail crane; The basic trajectory and the extended trajectory are recursively scanned to obtain the dual similarity measure of the monorail crane; Subband aggregation is performed on the dual similarity measure to obtain the frequency band entropy value set of the monorail crane.

8. The modularly integrated mine monorail control system as described in claim 1, characterized in that, When the traction force dynamic allocation scheme and multi-unit self-healing strategy generation module performs fault unit decoupling on the diagnostic primitives based on the fault rule base to obtain the traction force dynamic allocation scheme and multi-unit self-healing strategy for the monorail, it is specifically used for: The diagnostic primitives are subjected to feature dimensionality reduction to obtain the independent fault index of the monorail crane; The energy flow path is reconstructed for the independent fault index to obtain a dynamic traction force allocation view; Multimodal fusion is performed on the dynamic traction force distribution view to obtain the adaptive distribution matrix of the monorail. By redirecting the bus signals of the adaptive allocation matrix, the multi-link self-healing strategy of the monorail crane is obtained.

9. The modularly integrated mine monorail control system as described in claim 1, characterized in that, When the joint control signal generation module performs instruction fusion of the dynamic traction force distribution scheme and the multi-link self-healing strategy to obtain the joint control signal of the monorail main control core, and wirelessly transmits the joint control signal to the monorail main control core, it is specifically used for: The traction dynamic distribution scheme and the multi-link self-healing strategy are timestamped to obtain a synchronous command stream. The control field is filled into the synchronization command stream to obtain the joint control signal frame of the monorail crane; The control signal frame of the monorail is modulated and transmitted to obtain the radio frequency signal of the monorail. The radio frequency signal is spatially guided and propagated to obtain the main control core execution command of the monorail crane.

10. A control method for a mine monorail crane based on modular integration, characterized in that, The method includes: S1. Spatiotemporal fusion of multi-source heterogeneous data of the monorail crane is performed to obtain the operating state matrix of the monorail crane; S2. Extract fault features from the operating state matrix to obtain the abnormal feature vector of the monorail crane; S3. Perform adversarial spectrum reconstruction on the abnormal feature vector to obtain the vibration spectrum of the drive unit of the monorail crane; S4. Based on the load weight signal and real-time battery status signal of the monorail, perform wavelet packet sample entropy decomposition on the vibration spectrum of the drive unit to obtain the diagnostic primitives of the monorail. S5. Based on the fault rule base, the diagnostic primitives are decoupled from the fault units to obtain the dynamic traction force distribution scheme and multi-link self-healing strategy of the monorail. S6. The dynamic traction force distribution scheme and the multi-link self-healing strategy are fused together to obtain the joint control signal of the monorail main control core, and the joint control signal is wirelessly transmitted to the monorail main control core.

Images