A mine shovel transmission system predictive maintenance method and system based on electric vibration signal fusion

Abstract

This invention discloses a predictive maintenance method and system for a mining electric shovel transmission system based on electro-vibration signal fusion. The method includes: synchronously acquiring the three-phase current signal of the mining electric shovel's hoisting motor and the vibration acceleration signal of key monitoring points in its transmission system with microsecond-level timestamp accuracy; processing these signals to obtain a current spectrum feature vector and a vibration stress wave feature vector; concatenating the two vectors to obtain a joint feature vector; inputting the joint feature vector into a pre-trained health assessment model to output a continuous health index (HI) characterizing the real-time health status of the transmission system; and generating predictive maintenance decisions based on the HI. This invention, through the fusion analysis of motor current and mechanical vibration signals, achieves cross-validation and complementary enhancement of fault symptoms from both electrical and mechanical physical dimensions, significantly improving the detection sensitivity and reliability of early latent faults such as gear pitting and bearing wear.

Description

Technical Field

[0001] This invention belongs to the fields of equipment failure prediction and health management (PHM) and industrial Internet of Things, specifically relating to a predictive maintenance method and system for a mining electric shovel transmission system based on electro-vibration signal fusion. Background Technology

[0002] Electric shovels are core loading equipment in open-pit mining. Their hoisting and transmission systems operate under extremely harsh environments of heavy loads, strong impacts, and rapid changes in working conditions, making them a high-risk area for equipment failure. Traditional maintenance methods (such as planned maintenance and reactive repair) and single-dimensional condition monitoring technologies (such as vibration analysis or motor current analysis only) face fundamental challenges when applied to this specific scenario, mainly because existing technical solutions fail to adapt to the special working conditions of the electric shovel's transmission system. First, heavy loads and strong impacts frequently lead to signal distortion and false degradation. During electric shovel excavation, when the bucket encounters large pieces of material or incompletely blasted rock, the transmission system experiences a sudden and enormous torque impact, causing severe non-stationary distortion of the motor current signal. Conventional low-order harmonic indicators such as total harmonic distortion (THD) exhibit large oscillations unrelated to actual faults due to load fluctuations. Simultaneously, the vibration signal is filled with "pseudo-impact" components caused by load impacts rather than component damage, leading to frequent false alarms from traditional impact indicators such as kurtosis. Existing fusion methods designed for steady-state conditions (such as end-to-end fusion based on wavelet packet energy entropy or convolutional neural networks) lack anti-interference mechanisms for strong impact conditions. Directly adopting these methods not only fails to reduce false alarms but also causes a surge in misdiagnosis rates due to the confusion between load impacts and actual fault characteristics.

[0003] Secondly, frequent forward and reverse rotations and intermittent stalls lead to misalignment of characteristic times. An electric shovel typically completes one digging cycle in only 30-45 seconds, involving multiple stages such as bucket descent, digging and cutting, heavy-load lifting, and rotary unloading. The transmission system frequently experiences start-stop, forward and reverse rotations, and even stalls. This characteristic results in a physical time delay (approximately 0.5-2 ms) between the excitation and response of the current signal (reflecting changes in electromagnetic torque) and the vibration signal (reflecting mechanical impact response) to the same fault event. Without precise time-domain alignment, multi-source information fusion will lose its physical basis, leading to distorted fusion results.

[0004] Third, the rigid direct connection between the motor and the gearbox leads to strong coupling of fault modes. The electric shovel's lifting motor and gearbox are typically rigidly mechanically connected without any buffering mechanism. When pitting occurs in the gears or wear occurs in the bearings, minute changes in mechanical impedance are directly fed back to the motor's air gap magnetic field through the shaft, causing current harmonic distortion. Conversely, torque pulsations caused by electromagnetic faults in the motor also exacerbate gear meshing impacts. This "strong electromechanical coupling" characteristic makes the early fault characteristics of a single signal source extremely weak, easily overwhelmed by the strong signals of normal operation. Existing general fusion methods mostly focus on backend algorithm optimization and do not design feature layer fusion strategies specifically for this physical mechanism, resulting in insufficient sensitivity for early fault identification.

[0005] Fourth, the dispersed deployment of equipment limits data transmission and real-time analysis. Large open-pit mines typically deploy dozens of electric shovels, which are scattered and in mobile operation. If a centralized processing model is adopted, uploading all raw high-frequency current and vibration waveform data to the cloud, the bandwidth of existing industrial wireless networks simply cannot support it, making it difficult to scale up the solution.

[0006] Furthermore, in the field of electric shovel fault diagnosis, there is a common technical bias among those skilled in the art: they believe that harmonic distortion under strong impact loads cannot be used for fault diagnosis because the impact itself causes violent power fluctuations. Traditionally, time-domain characteristics or low-order harmonics are preferred, thus forming the inherent perception that "high-order harmonics are not applicable to impact conditions".

[0007] In summary, existing technical solutions, when directly applied to the drive system of electric shovels in mines, face problems such as signal failure, inaccurate fusion, and implementation difficulties due to the aforementioned special working conditions. Therefore, there is an urgent need for a predictive maintenance method and system that can fundamentally adapt to these special working conditions and achieve reliable, accurate, and engineerable deployment. Summary of the Invention

[0008] To overcome the numerous applicability difficulties of existing technologies, as mentioned in the background, due to their inability to adapt to the special working conditions of mining electric shovel transmission systems, this invention provides a predictive maintenance method and system for mining electric shovel transmission systems based on electro-vibration signal fusion. This method not only performs multi-source signal fusion but, more importantly, designs a complete and targeted solution based on the unique working characteristics of electric shovel transmission systems, such as "heavy load and strong impact, rapid cycle, strong electro-mechanical coupling, and distributed deployment." This solution addresses the passivity and inapplicability of traditional methods, driving the leap from traditional operation and maintenance to intelligent predictive maintenance.

[0009] A predictive maintenance method for a mine electric shovel drive system based on electro-vibration signal fusion, the method comprising the following steps: H1. The three-phase current signal of the hoisting motor of the mining electric shovel and the vibration acceleration signal of the key monitoring points of its transmission system are synchronously collected with microsecond-level timestamp accuracy; H2. The three-phase current signal is processed to obtain the current spectrum feature vector, and the vibration acceleration signal is processed to obtain the vibration stress wave feature vector; H3. Concatenate the current spectrum feature vector with the vibration stress wave feature vector to obtain a joint feature vector; H4. Input the joint feature vector into the pre-trained health assessment model and output a continuous health index HI that represents the real-time health status of the transmission system; H5. Generate predictive maintenance decisions based on the continuous health index.

[0010] In addition to the aspects and any possible implementations described above, an implementation is further provided in which H2 includes: H2.1. The three-phase current signal is subjected to Clark-Parker transform to obtain the quadrature-axis torque current component. The spectrum analysis of this component is performed to remove the third and lower harmonics that are significantly affected by load fluctuations. The total distortion rate of the fourth and higher harmonics is calculated, and the ratio of the amplitude of the fifth harmonic to the amplitude of the fundamental wave and the ratio of the amplitude of the seventh harmonic to the amplitude of the fundamental wave are extracted to form the current spectrum feature vector. The fifth and seventh harmonics are sensitive to the electromagnetic asymmetry of the motor and load, and are robust to pure load fluctuations. H2.2. Vibration stress wave feature extraction: The vibration acceleration signal is subjected to high-pass filtering to extract the time-domain peak value and kurtosis of the filtered signal; the filtered signal is then envelope demodulated to obtain the envelope signal, and the envelope signal is subjected to spectrum analysis to extract the amplitude at the bearing fault characteristic frequency to form the vibration stress wave feature vector.

[0011] In addition to the aspects described above and any possible implementation, a further implementation is provided in which the continuous health index HI ranges from 0 to 100.

[0012] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the extraction method of the q-axis torque current component in H2.1 is as follows: first, the current in the three-phase stationary coordinate system is converted into the current in the two-phase stationary coordinate system through the Clark transformation, and then the current in the two-phase stationary coordinate system is converted into the quadrature-axis current in the synchronous rotating coordinate system through the Park transformation; this component directly corresponds to the electromagnetic torque output by the motor and has a physical mapping relationship with the mechanical torque fluctuation of the transmission system.

[0013] In addition to the aspects described above and any possible implementation, a further implementation is provided in which the bearing failure characteristic frequency in H2.2 is calculated using the formula BPFI = (Z / 2) * (n / 60) * (1+(d / D) * cosα), where Z is the number of bearing rollers, n is the shaft speed, d is the roller diameter, D is the bearing pitch circle diameter, and α is the roller contact angle.

[0014] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the health assessment model is a regression model based on gradient boosting decision trees, and the training samples are joint feature vectors of the same model of electric shovel under different health states and corresponding manually labeled health index labels, covering multiple states such as healthy, early pitting corrosion, moderate wear and severe failure.

[0015] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein H5 specifically includes: H5.1. Analyzing the historical time series data of the HI: After filtering the continuous health index using the 5-point median to eliminate instantaneous fluctuations, it is determined whether the continuous health index is monotonically decreasing within 5 consecutive sampling periods and the cumulative decrease is ≥5%. If it is satisfied, a warning of the corresponding level is triggered; and at the same time, the filtered continuous health index sequence is fitted with linear least squares to obtain the degradation rate, which is extrapolated to a preset maintenance threshold to predict and improve the remaining service life of key system components, such as gears and bearings; H5.2. Generate predictive maintenance decisions based on early warning information, remaining service life prediction results, and spare parts inventory.

[0016] In addition to the aspects described above and any possible implementations, a further implementation is provided in which the predictive maintenance decision includes maintenance priority, window period, and spare parts replacement recommendations.

[0017] In addition to the aspects described above and any possible implementation, a further implementation is provided in which the cumulative reduction is calculated as ((first filter HI - last filter HI) / first filter HI) × 100%.

[0018] The present invention also provides a predictive maintenance system for a mine electric shovel transmission system based on electro-vibration signal fusion. The system is used to implement the method and includes the following modules: The sensing and acquisition module includes at least one three-phase current sensor and at least one vibration acceleration sensor, used to synchronously acquire the three-phase current signal of the hoisting motor of the mining electric shovel and the vibration acceleration signal of the key monitoring points of its transmission system with microsecond-level timestamp accuracy; The edge intelligent computing module is communicatively connected to the sensing and acquisition module. It is used to process the three-phase current signal to obtain the current spectrum feature vector, process the vibration acceleration signal to obtain the vibration stress wave feature vector, and concatenate the current spectrum feature vector and the vibration stress wave feature vector to obtain a joint feature vector and output it. The cloud-based data analysis and decision-making module is connected to the edge intelligent computing module. It is used to train the received joint feature vector using a pre-trained health assessment model to obtain a continuous health index that characterizes the real-time health status of the transmission system. Based on the health index, it obtains health assessment results, early warning information, prediction results, and maintenance decisions. The human-computer interaction and display module communicates with the cloud data analysis and decision-making module to visualize and interact with health assessment results, early warning information, prediction results, and maintenance decisions.

[0019] Beneficial effects of the present invention This invention, through the fusion and intelligent analysis of electro-vibration signals, effectively overcomes the shortcomings of the single-signal monitoring schemes described in the background art, such as weak signals and insufficient specificity in early-stage faults, as well as the passive nature of traditional maintenance methods, achieving the following beneficial effects: 1) Early Fault Detection Capability Based on Multi-Source Information Fusion: By fusing and analyzing motor current and mechanical vibration signals, cross-verification and complementary enhancement of fault symptoms from both electrical and mechanical physical dimensions are achieved. This directly overcomes the shortcomings of previous technologies where single signal sources produce weak or insufficiently specific signals in early faults, thus significantly improving the detection sensitivity and reliability of early latent faults such as gear pitting and bearing wear, achieving "early detection" of faults.

[0020] 2) Intelligent assessment and accurate diagnosis capabilities based on fused features: The joint feature vector formed by fusing current spectrum features and vibration stress wave features provides a more comprehensive and complementary information dimension for machine learning models. By learning from these fused deep features, the model can more effectively distinguish between normal operating condition fluctuations and true early fault buds, and make more accurate preliminary identifications of fault types, thereby significantly reducing the false alarm rate and achieving "accurate judgment" of equipment health status.

[0021] 3) Forward-looking decision-making based on trend forecasting: It not only focuses on the current state, but also pays attention to the historical trend of health assessment index. Through trend extrapolation and degradation modeling, it can predict the remaining service life of components, providing a direct basis for formulating economical and scientific maintenance plans and realizing "early planning" of maintenance.

[0022] 4) Advanced system architecture and strong practicality: The method has a clear logic, and the corresponding system can adopt a collaborative architecture of "edge-side feature extraction and fusion + cloud-based intelligent analysis and decision-making". It has good compatibility with the transformation of existing equipment, low data transmission bandwidth requirements, and powerful analysis and early warning capabilities, and has good engineering application and promotion value. Attached Figure Description

[0023] Figure 1 This is a system architecture and data flow diagram of the present invention; Figure 2 This is the overall flowchart of the predictive maintenance method of the present invention. Detailed Implementation

[0024] To better understand the technical solution of this invention, the content of this invention includes, but is not limited to, the specific embodiments described below. Similar technologies and methods should be considered within the scope of protection of this invention. To make the technical problems to be solved, the technical solutions, and advantages of this invention clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments.

[0025] It should be understood that the embodiments described in this invention are merely some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0026] This invention breaks away from conventional multi-source fusion thinking and constructs a customized link for the unique four constraints of electric shovels: To address the false degradation of signals under heavy loads and strong impacts, it abandons the conventional THD index, which is easily affected by load fluctuations, and innovatively selects the 5th / 7th harmonics, which are sensitive to electromagnetic asymmetry and robust to pure load fluctuations, as current characteristics, fundamentally cutting off load interference; to address time delay misalignment under rapid changing cycles, it uses microsecond-level synchronous sampling to lock the physical time, ensuring consistent fusion benchmarks; to address the weak characteristics under strong electromechanical coupling, it uses q-axis current as an "electromagnetic probe" to perform physical cross-verification with vibration stress waves, solving the problem of single signals being unable to decouple; and to address the transmission bottleneck of distributed deployments, it pushes the feature fusion with high computing power requirements down to the edge.

[0027] This invention provides a predictive maintenance method for the transmission system of a mining electric shovel based on electro-vibration signal fusion. Designed for the special working conditions of mining electric shovel transmission systems, including heavy loads and strong impacts, rapid changes in operating cycles, rigid direct connection between the motor and gearbox, and dispersed equipment deployment, the method includes the following steps: S1: Multi-source signal microsecond-level synchronous acquisition: The three-phase current signal of the hoisting motor of the mining electric shovel and the vibration acceleration signal of the transmission system are synchronously acquired with microsecond-level timestamp accuracy through edge acquisition equipment; Key monitoring points include the bearings at the drive and non-drive ends of the motor, the input and output ends of the coupling, and the input and output bearings of the gearbox. Monitoring these locations is crucial for improving the axial and radial vibration acceleration signals of the drum bearings, etc. To meet the quantitative relationship between phase error and time deviation and to diagnose faults such as coupling misalignment, all vibration signal sampling data are labeled with a timestamp deviation of less than 100ns, thereby improving the system's dual motors and transmission system are connected by rigid coupling.

[0028] S2: Cross-domain feature extraction and anti-interference fusion: S2.1 Impact Current Spectrum Feature Extraction: The three-phase current signal is subjected to Clark-Parker transform to obtain the quadrature axis (q-axis) torque current component. Spectral analysis is performed on this component to remove low-order harmonics of the 3rd order and below that are significantly affected by load fluctuations. The total distortion rate of harmonics of the 4th order and above is calculated. The ratios of the amplitude of the 5th harmonic to the fundamental amplitude and the amplitude of the 7th harmonic to the fundamental amplitude are extracted to form a current spectrum feature vector. This can filter out load artifacts, retain true fault characteristics, and significantly reduce false alarms under strong impact conditions. The 5th and 7th harmonics have a significant impact on the electromagnetic incompatibility of the motor and load. It is sensitive to symmetry and robust to pure load fluctuations. The method for obtaining the q-axis torque current component is as follows: first, the current in the three-phase stationary coordinate system is converted into the current in the two-phase stationary coordinate system through the Clark transformation, and then the current in the two-phase stationary coordinate system is converted into the q-axis current in the synchronous rotating coordinate system through the Park transformation. This technology is a mature mathematical model, and will not be elaborated further in this invention. The q-axis component directly corresponds to the electromagnetic torque output by the motor and has a physical mapping relationship with the mechanical torque fluctuation of the transmission system. Under the condition of rigid direct connection between the electric shovel motor and the gearbox, this mapping relationship is not affected by the flexible buffer.

[0029] S2.2 Vibration Stress Wave Feature Extraction: The vibration acceleration signal is high-pass filtered to extract the time-domain peak value and kurtosis of the filtered signal, which are directly used as components of the vibration stress wave feature vector. This vector is concatenated with the current spectrum feature vector to form a joint feature vector, which is input into the pre-trained gradient boosting decision tree (GBDT) regression model. The model uses these two features to perform health state regression and outputs a continuous health index (HI). The filtered signal is envelope demodulated to obtain an envelope signal. The envelope signal is then subjected to spectral analysis to extract the amplitude at the bearing fault characteristic frequency of the transmission system, forming the vibration stress wave feature vector. The bearing fault characteristic frequency of the transmission system is calculated using the formula BPFI=(Z / 2)* (n / 60) * (1+(d / D) * cosα), where Z is the number of bearing rollers, n is the shaft speed, d is the roller diameter, D is the bearing pitch circle diameter, and α is the roller contact angle.

[0030] S2.3 Explicit Feature Layer Fusion: The current spectrum feature vector and the vibration stress wave feature vector are concatenated to obtain a joint feature vector; the current spectrum feature reflects the mechanical impedance change sensed by the motor side, and the vibration stress wave feature reflects the impact event that occurs in the mechanical body. The two form a cross-verification of the electro-mechanical coupling fault. S3: Health Status Assessment and Prediction S3.1 The joint feature vector is input into the pre-trained health assessment model, which outputs a continuous health index HI representing the real-time health status of the transmission system. The value of HI ranges from 0 to 100, with higher values ​​indicating better health. The health assessment model is a regression model based on a gradient boosting decision tree. The training samples are the joint feature vectors of the same model of electric shovel under different health states and their corresponding manually labeled health index tags, covering healthy conditions and early pitting corrosion, such as damage areas <2mm. 2 It belongs to the very early stage of microscopic damage, which has almost no impact on the operation of the equipment macroscopically, but there are already weak high-frequency stress waves and electromagnetic harmonic anomalies, moderate wear, and severe failure. Due to the clear physical meaning and shock resistance of the input features, this model has better engineering interpretability and small sample generalization ability compared to neural networks.

[0031] S3.2 Analyze based on the historical time series data of HI: After eliminating instantaneous fluctuations by using 5-point median filtering, determine whether the HI monotonically decreases within 5 consecutive sampling periods and the cumulative decrease is ≥5%. If it is satisfied, trigger the corresponding level of warning; use linear least squares fitting to obtain the degradation rate of the filtered HI sequence and extrapolate it to the preset maintenance threshold. HI=30, predicted remaining service life; The warning logic is as follows: After applying a 5-point median filter to the HI time-series data, it is determined whether the HI value strictly decreases monotonically and the cumulative decrease is ≥5% within 5 consecutive sampling periods, where the cumulative decrease = ((first filtered HI - last filtered HI) / first filtered HI) × 100%. If this condition is met, a warning is issued. The remaining lifetime prediction uses the linear least squares method to fit the most recent 20 filtered HI values ​​to obtain the degradation rate, in units of HI / day. Specifically, it is formed by a sliding window of the 20 most recent consecutive health index sampling points that have undergone 5-point median filtering, traced back from the current moment. Selecting the most recent 20 data points is to track the current degradation trend in a timely manner and avoid using old data, such as stable periods from several months ago, to flatten the degradation rate and cause a lag in remaining lifetime prediction.

[0032] Based on engineering experience, when the health index drops to 30, it indicates that approximately 70% of the transmission system's performance reserves have been depleted, and the inflection point of accelerated deterioration is approaching, but functional failure has not yet occurred. Therefore, this invention extrapolates to a maintenance threshold HI=30 to predict the remaining effective days. Specifically, the degradation rate k is obtained by fitting the most recent 20 filtered HI values ​​using the linear least squares method, with units of HI / day. The current filtered HI is HI_current, the maintenance threshold HI_th=30, and the extrapolation formula for the remaining effective days RUL is: RUL=(HI_current-30) / |k|, which is effective when k<0.

[0033] S4: Predictive maintenance decision generation: Based on real-time dynamic analysis of HI, automatically generated early warning information, remaining service life prediction results, and spare parts inventory, predictive maintenance plans are generated, including maintenance priorities, window periods, and spare parts replacement recommendations.

[0034] This invention adopts a cloud-edge-device collaborative architecture, in which the edge device is deployed at the electric shovel site to complete synchronous signal acquisition, feature extraction and fusion. Only low-dimensional joint features, such as vector dimensions ≤6, are uploaded to the cloud via 4G / 5G industrial wireless network to adapt to the bandwidth constraints of mining industrial wireless networks. When the network is interrupted, the edge device temporarily stores the feature data of the most recent 72 hours and automatically retransmits it after the network is restored.

[0035] To further verify the technical effect of the present invention, a comparative test was conducted on the WK-35 electric shovel simulation platform, and the results are shown in Table 1: Table 1 Test Results Comparison Table

[0036] The test results show that the present invention has significant advantages over the prior art in terms of detection rate, false alarm rate and early warning lead time, producing unexpected technical effects.

[0037] The following specific examples illustrate this.

[0038] Example 1: A predictive maintenance method for the drive system of WK series electric shovels.

[0039] This embodiment uses the WK-35 electric shovel hoisting system of Dexing Copper Mine as an example. Figure 1 As shown, an open-type closed-loop Hall current sensor is installed on the power supply busbar of the hoisting motor for non-invasive acquisition of three-phase current; an IEPE type vibration acceleration sensor is installed vertically on the input shaft bearing housing and the output shaft bearing housing of the hoisting gearbox. All sensor signals are connected to an edge intelligent data acquisition terminal deployed in the electric shovel's electrical room via shielded cables.

[0040] S1: Synchronous Signal Acquisition: The edge intelligent acquisition terminal synchronously acquires three-phase current signals at a sampling rate of 2kHz and two vibration acceleration signals at a sampling rate of 10kHz. The acquisition terminal has a built-in high-precision clock that timestamps all sampled data points with microsecond-level precision (1μs resolution). Since the electric shovel frequently changes direction within a 30-45 second cycle, the physical response delay between the current sensor and the vibration sensor is usually between 0.5-2ms. Using microsecond-level timestamps can completely eliminate this difference, ensuring strict synchronization of multi-source signals and laying the foundation for subsequent fusion analysis.

[0041] S2: Feature Extraction and Fusion: The edge acquisition terminal has online computing capabilities, performing feature calculations on the cached data every 10 seconds to generate a set of joint feature vectors. 1. Current spectrum feature extraction: The collected instantaneous three-phase current values ​​were subjected to Clark-Parker transform to obtain the q-axis current component representing the electromagnetic torque of the motor. A 2048-point Fast Fourier Transform (frequency resolution approximately 0.98Hz) was performed on the sequence to obtain the amplitude A1 of the 50Hz power frequency fundamental wave from the spectrum.

[0042] (1) Total distortion rate of 4th and higher harmonics: To avoid load fluctuation interference, low-order harmonics (150Hz and below) of the 3rd order are removed from the FFT spectrum, and only components of the 4th order and above are considered. The calculation formula is as follows: Where A1 is the fundamental amplitude, A k The amplitude of the fourth harmonic or higher.

[0043] (2) Relative amplitude of specific harmonics: Calculate the ratios H5 / Fund and H7 / Fund of the amplitude of the 5th harmonic (250Hz) and the amplitude of the 7th harmonic (350Hz) to the fundamental amplitude. These harmonics are related to the electromagnetic asymmetry of the motor and load, cogging torque pulsation, etc., and can sensitively reflect the changes in the mechanical state of the transmission system, but are not sensitive to pure load fluctuations.

[0044] (3) Generate current spectrum feature vector: F_current=[THD, H5 / Fund, H7 / Fund].

[0045] 2. Vibration stress wave feature extraction: The original vibration acceleration signal is subjected to a 2kHz high-pass digital filter (Butterworth, 4th order) to suppress low-frequency structural vibration interference (electric shovel body sway is usually below 20Hz) and highlight the high-frequency impact components (usually concentrated in 2-8kHz) generated by bearing and gear defects.

[0046] (1) Time-domain indicators: Peak and kurtosis of the original vibration signal. The kurtosis value is extremely sensitive to impact signals.

[0047] (2) Envelope demodulation: The filtered signal is subjected to Hilbert transform to obtain the analytic signal, and the envelope signal is obtained by taking the modulus. The envelope spectrum is obtained by performing a 1024-point FFT on the envelope signal (frequency resolution of approximately 9.8 Hz). (3) Frequency domain index: Extract the amplitude Env_BPFI at the fault characteristic frequency of the inner ring of the input shaft bearing from the envelope spectrum. BPFI can be calculated based on the bearing model parameters and shaft speed: BPFI = (Z / 2) * (n / 60) * (1+ (d / D) * cosα).

[0048] Therefore, the generated vibration stress wave eigenvector is: F_vibration=[Peak, Kurtosis,Env_BPFI].

[0049] 3. Feature fusion: The current spectrum feature vector and the vibration stress wave feature vector are concatenated to obtain the joint feature vector: F_joint=[THD, H5 / Fund, H7 / Fund, Peak, Kurtosis, Env_BPFI].

[0050] The joint feature vector has a dimension of 6, which is much smaller than the original waveform data (the original 10kHz sampled data for 10 seconds has 100,000 points), thus achieving data "slimming".

[0051] The current spectrum features reflect the changes in mechanical impedance sensed by the motor side, and the vibration stress wave features reflect the impact events that occur in the mechanical body. The explicit splicing of the two at the feature layer forms a cross-verification in a physical sense, which is specifically used to reveal early composite faults under the "strong electro-mechanical coupling" characteristics of the electric shovel transmission system.

[0052] S3: Health Assessment and Prediction.

[0053] The edge acquisition terminal compresses the joint feature vector generated every 10 seconds and uploads it to the cloud-based predictive maintenance platform server via a 4G industrial network.

[0054] 1. Health Assessment: A pre-trained gradient boosting decision tree regression model is deployed in the cloud as the health assessment model. The gradient boosting decision tree model was chosen over a neural network model because field samples of electric shovels are scarce and fault labels are extremely difficult to obtain. This model exhibits stronger robustness and generalization ability when handling small-sample, high-dimensional industrial feature data. The model is trained using historical data from a year of operational monitoring of multiple electric shovels of the same model, and the corresponding health status has been labeled by experts (covering healthy and early bearing pitting corrosion (damage area <2mm)). 2 (Multiple states including moderate gear wear and severe failure). After receiving the joint feature vector, the model outputs a health index HI between 0 and 100. HI=100 represents the ideal health state, and HI=0 represents functional failure.

[0055] 2. Early warning and prediction: The platform continuously stores and analyzes the time series of HI.

[0056] (1) Early warning logic: First, the HI time series data is filtered using a 5-point median to eliminate instantaneous fluctuations caused by load impact. The early warning criterion is set as follows: the HI value strictly decreases monotonically within 5 consecutive sampling periods (i.e., 50 seconds) after filtering, and the cumulative decrease is ≥5%, where the cumulative decrease = ((first filtered HI - last filtered HI) / first filtered HI) × 100%. The threshold is set based on field measurement data showing that the HI fluctuation caused by pure load fluctuation is usually within ±2%, while the degradation trend caused by early faults is monotonic and cumulative. Setting a 5% threshold can effectively filter out excavation impact interference and achieve accurate early warning.

[0057] (2) Lifespan Prediction: The linear least squares method is used to fit the most recent 20 filtered HI values ​​to obtain the degradation rate (unit: HI / day). The trend is then extrapolated to a preset maintenance threshold HI=30 to predict the remaining effective working days RUL. The calculation formula is: RUL= (30-HI_current) / degradation rate, where HI_current is the current filtered HI value. The predicted remaining effective working days are used for planned shutdown maintenance decisions.

[0058] S4: Predictive Maintenance Decision Generation: The cloud platform automatically generates maintenance plans based on the warning level, remaining life prediction results, and spare parts inventory database. These plans include: maintenance priority (high / medium / low); suggested maintenance windows (e.g., within the next 7 days, 15 days, or 30 days); and a spare parts replacement list (specific models and quantities). This plan is then pushed to equipment administrators and maintenance engineers via a human-machine interface.

[0059] Example 2: A predictive maintenance system.

[0060] like Figure 1 As shown, this system adopts a multi-layered architecture that integrates cloud, edge, and device to achieve a closed loop from data collection to intelligent decision-making. 1. End-side sensing layer: Composed of the aforementioned current sensor and vibration acceleration sensor, it is responsible for sensing the original physical signals.

[0061] 2. Edge Preprocessing and Edge Computing Layer: Composed of edge intelligent acquisition terminals, its core task is to "slim down" the data. To this end, it is responsible for high-speed synchronous acquisition of multiple signals, filtering, feature extraction and other real-time computations, and only uploads the 6-dimensional joint feature vector, which greatly reduces the network transmission bandwidth pressure. It can also temporarily store the feature data of the most recent 72 hours when the network is interrupted and automatically re-upload it after the network is restored.

[0062] 3. Cloud Platform Layer: Composed of a cluster of cloud computing servers. It is responsible for storing and managing massive amounts of historical data, running complex health assessment and prediction algorithm models, performing in-depth analysis and mining across devices and time dimensions, and realizing advanced functions such as centralized alarm management, automatic report generation, and maintenance work order workflow.

[0063] 4. Application Interaction Layer: Provides a visual interface to equipment administrators and maintenance engineers via a web browser or mobile app. The interface can dynamically display key information such as real-time HI value, historical trend curves, current warning list, and predicted remaining effective working time, and supports interactive query and analysis.

[0064] As an embodiment of the present invention, the present invention also discloses a predictive maintenance system for implementing the method, comprising: The sensing and acquisition module, deployed on the device side, includes at least one three-phase current sensor and at least one vibration acceleration sensor, for synchronously sampling current and vibration signals at the microsecond level; The edge intelligent computing module is deployed on the equipment site, i.e., the electric shovel working site. It is communicatively connected to the sensing and acquisition module, and has a built-in real-time operating system and feature extraction algorithm. It is used to execute the S2 step, output the joint feature vector, and upload it through the 4G / 5G module. The cloud-based data analysis and decision-making module, deployed in the data center, has main functions including offline model training and iterative optimization, real-time status monitoring and digital dashboards, fault snapshot storage and accident backtracking, remote expert consultation and waveform playback, OTA firmware and parameter upgrade management, etc. It communicates with the edge intelligent computing module and has a pre-trained health assessment model embedded in it to execute the S3 steps and output health index, early warning and remaining life prediction. The human-computer interaction and display module communicates with the cloud data analysis and decision-making module and is used to visualize and interact with health assessment results, early warning information, prediction results and maintenance suggestions.

[0065] 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 “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0066] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A predictive maintenance method for a mine electric shovel transmission system based on electro-vibration signal fusion, characterized in that, The method includes the following steps: H1. The three-phase current signal of the hoisting motor of the mining electric shovel and the vibration acceleration signal of the key monitoring points of its transmission system are synchronously collected with microsecond-level timestamp accuracy; H2. The three-phase current signal is processed to obtain the current spectrum feature vector, and the vibration acceleration signal is processed to obtain the vibration stress wave feature vector; H3. Concatenate the current spectrum feature vector with the vibration stress wave feature vector to obtain a joint feature vector; H4. Input the joint feature vector into the pre-trained health assessment model and output a continuous health index HI that represents the real-time health status of the transmission system; H5. Generate predictive maintenance decisions based on the continuous health index.

2. The method according to claim 1, characterized in that, The H2 includes: H2.

1. The three-phase current signal is subjected to Clark-Parker transform to obtain the quadrature-axis q-axis torque current component. The spectrum analysis of this component is performed to remove the third and lower harmonics that are significantly affected by load fluctuations. The total distortion rate of the fourth and higher harmonics is calculated, and the ratio of the amplitude of the fifth harmonic to the amplitude of the fundamental wave and the ratio of the amplitude of the seventh harmonic to the amplitude of the fundamental wave are extracted to form the current spectrum feature vector. The fifth and seventh harmonics are sensitive to the electromagnetic asymmetry of the motor and load, and are robust to pure load fluctuations. H2.

2. Vibration stress wave feature extraction: The vibration acceleration signal is subjected to high-pass filtering to extract the time-domain peak value and kurtosis of the filtered signal; the filtered signal is then envelope demodulated to obtain the envelope signal, and the envelope signal is subjected to spectrum analysis to extract the amplitude at the bearing fault characteristic frequency to form the vibration stress wave feature vector.

3. The method according to claim 1, characterized in that, The continuous health index HI ranges from 0 to 100.

4. The method according to claim 2, characterized in that, The extraction method of the q-axis torque current component in H2.1 is as follows: first, the current in the three-phase stationary coordinate system is converted into the current in the two-phase stationary coordinate system through the Clark transformation, and then the current in the two-phase stationary coordinate system is converted into the quadrature-axis current in the synchronous rotating coordinate system through the Park transformation; this component directly corresponds to the electromagnetic torque output by the motor and has a physical mapping relationship with the mechanical torque fluctuation of the transmission system.

5. The method according to claim 1, characterized in that, The bearing failure characteristic frequency in H2.2 is calculated using the formula BPFI=(Z / 2) * (n / 60) * (1+(d / D)*cosα), where Z is the number of bearing rollers, n is the shaft speed, d is the roller diameter, D is the bearing pitch circle diameter, and α is the roller contact angle.

6. The method according to claim 1, characterized in that, The health assessment model is a regression model based on gradient boosting decision tree. The training samples are the joint feature vectors of the same model of electric shovel under different health states and the corresponding manually labeled health index labels, covering multiple states such as healthy, early pitting corrosion, moderate wear and severe failure.

7. The method according to claim 1, characterized in that, The H5 specifically includes: H5.

1. Based on the historical time series data of the HI, the continuous health index is filtered using the 5-point median to eliminate instantaneous fluctuations. It is then determined whether the continuous health index decreases monotonically within 5 consecutive sampling periods and the cumulative decrease is ≥5%. If it is satisfied, an early warning of the corresponding level is triggered. At the same time, the degradation rate is obtained by linear least squares fitting of the filtered continuous health index sequence and extrapolated to the preset maintenance threshold to predict the remaining service life of the key components of the improvement system. H5.

2. Generate predictive maintenance decisions based on early warning information, remaining service life prediction results, and spare parts inventory.

8. The method according to claim 1, characterized in that, The predictive maintenance decisions include maintenance priorities, window periods, and spare parts replacement recommendations.

9. The method according to claim 7, characterized in that, The formula for calculating the cumulative reduction is ((first filter HI - last filter HI) / first filter HI) × 100%.

10. A predictive maintenance system for a mine electric shovel transmission system based on electro-vibration signal fusion, characterized in that, The system is used to implement the method according to any one of claims 1-9, and includes the following modules: The sensing and acquisition module includes at least one three-phase current sensor and at least one vibration acceleration sensor, used to synchronously acquire the three-phase current signal of the hoisting motor of the mining electric shovel and the vibration acceleration signal of the key monitoring points of its transmission system with microsecond-level timestamp accuracy; The edge intelligent computing module is communicatively connected to the sensing and acquisition module. It is used to process the three-phase current signal to obtain the current spectrum feature vector, process the vibration acceleration signal to obtain the vibration stress wave feature vector, and concatenate the current spectrum feature vector and the vibration stress wave feature vector to obtain a joint feature vector and output it. The cloud-based data analysis and decision-making module is connected to the edge intelligent computing module. It is used to train the received joint feature vector using a pre-trained health assessment model to obtain a continuous health index that characterizes the real-time health status of the transmission system. Based on the health index, it obtains health assessment results, early warning information, prediction results, and maintenance decisions. The human-computer interaction and display module communicates with the cloud data analysis and decision-making module to visualize and interact with health assessment results, early warning information, prediction results, and maintenance decisions.

Images