A large energy forging hammer remote operation and maintenance system and method

Abstract

The application discloses a large-energy forging hammer remote operation and maintenance and maintenance system and method, relates to the technical field of industrial equipment operation and maintenance, and the fault diagnosis model library is based on a forging hammer digital twin body, simulates standard vibration frequency spectrum, stress distribution, motion trajectory and acoustic emission characteristics of a forging hammer fault type and a forging hammer fault degree, extracts time sequence characteristics of a forging hammer state parameter based on an LSTM, extracts spatial characteristics of the forging hammer state parameter based on a deep residual network, highlights a sensor channel sensitive to a fault by weighting sensor channels of a device perception layer, weights time sequence characteristics output by the LSTM, focuses on a time point at which a fault occurs, and combines feature collection of stress concentration areas and fault high-incidence areas of the forging hammer digital twin body to complete construction of a fault diagnosis model. The application is designed with a fault diagnosis model that is trained in cooperation with a forging hammer digital twin body sample and a real sample, and realizes the function of accurate diagnosis of forging hammer faults.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment operation and maintenance technology, specifically to a remote operation and maintenance system and method for a high-energy forging hammer. Background Technology

[0002] Forging hammers are used in high-end equipment manufacturing such as aerospace, automobile manufacturing and rail transportation. Among them, high-energy forging hammers are widely used in the production of large structural parts due to their high impact energy and high forming accuracy. However, these types of equipment work under harsh conditions of high frequency impact, high temperature and high dust for a long time, and their key components are very prone to fatigue damage and performance degradation.

[0003] The existing monitoring system for forging hammers can only provide alarm prompts for a single type of fault. It cannot simultaneously complete the precise location of the fault and the quantitative assessment of its severity. After the diagnostic results are output, maintenance personnel need to spend 3-4 hours conducting manual troubleshooting to determine the specific fault location and repair plan, which seriously affects production efficiency.

[0004] Patent CN113204220B discloses an operation and maintenance system for industrial equipment. The patent realizes the analysis and correlation of automatic operation logs generated by various equipment and routine operation and maintenance logs filled in manually with environmental data, providing reference data for the optimal environment for the operation and maintenance of different types of industrial equipment, and also providing a theoretical basis for the number of spare parts, equipment and consumables in the maintenance process.

[0005] The aforementioned patent collects operational status data of various industrial equipment through an industrial equipment operation data acquisition module, and collects environmental data of the location of one or more industrial equipment through an industrial equipment operation environment data acquisition module including multiple environmental data acquisition terminals. The data transmission module sends the operational status data of each industrial equipment and its corresponding environmental data to the processing module. The processing module sets the numerical range of various levels of environmental data for various types, records the correspondence between each type of industrial equipment and its environmental data level, and stores it in the data storage module. This not only enables real-time alarms for the operational status of industrial equipment through operational data status analysis, reducing human maintenance costs, but also allows the correlation between the operational status data and environmental data to provide basic data for the subsequent operation and maintenance environment of industrial equipment. However, there is room for optimization in terms of fault diagnosis accuracy.

[0006] Therefore, this application proposes a remote operation and maintenance system and method for accurately diagnosing forging hammer faults in high-energy forging hammers. Summary of the Invention

[0007] The purpose of this invention is to provide a remote operation and maintenance system and method for high-energy forging hammers, so as to solve the technical problems mentioned in the background art, which can only perform single fault alarms and cannot confirm the fault location and severity.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a remote operation and maintenance system for a high-energy forging hammer, comprising an equipment perception layer, an edge computing layer, and a cloud-based operation and maintenance platform layer. The equipment perception layer is used to collect forging hammer status parameters, the edge computing layer is used for real-time feature extraction and anomaly detection of the forging hammer status parameters, and the cloud-based operation and maintenance platform layer consists of a digital twin engine, a fault diagnosis model library, a life prediction module, a health assessment engine, and a maintenance decision optimizer.

[0009] The digital twin engine receives the forging hammer state parameters processed by the edge computing layer and constructs a digital twin of the forging hammer. The fault diagnosis model library, based on the forging hammer digital twin, simulates the standard vibration spectrum, stress distribution, motion trajectory, and acoustic emission characteristics of the forging hammer fault type and fault degree. It extracts the temporal features of the forging hammer state parameters based on LSTM and the spatial features based on deep residual networks. By weighting the sensor channels of the equipment perception layer, it highlights the sensor channels sensitive to faults. It also weights the temporal features output by LSTM to focus on the time point of fault occurrence. Combining the feature acquisition of stress concentration areas and high-incidence areas by the forging hammer digital twin, the fault diagnosis model is constructed. The fault diagnosis model outputs the probability distribution of fault type and fault location and evaluates the fault degree.

[0010] Preferably, the device sensing layer consists of a vibration sensing unit, a temperature sensing unit, a pressure sensing unit, a displacement and deformation sensing unit, an acoustic sensing unit, and an energy consumption monitoring unit.

[0011] The vibration sensing unit consists of a triaxial accelerometer and a piezoelectric accelerometer, which are deployed at the hammer, anvil, guide rail and bearing to collect impact energy, rebound characteristics, guidance accuracy and bearing condition.

[0012] The temperature sensing unit consists of a PT1000 temperature sensor and an infrared thermal imager, and is deployed in the hydraulic system, motor bearings, gearbox and anvil to collect temperature distribution and temperature rise trend.

[0013] The pressure sensing unit is a high-frequency piezoresistive pressure sensor, which is deployed in the main oil circuit, accumulator and impact cylinder of the hydraulic system to collect hydraulic impact waveforms, pressure fluctuations and sealing status.

[0014] The displacement and deformation sensing unit consists of a laser displacement sensor and a resistance strain gauge, which is deployed on the guide rail to collect hammer stroke, anvil sinkage and frame deformation.

[0015] The acoustic sensing unit, consisting of a microphone array and an acoustic emission sensor, is deployed at the hammerhead for internal crack identification and cavitation detection.

[0016] The energy consumption monitoring unit consists of a smart meter and a harmonic analyzer, and is deployed in the electrical control cabinet to collect motor current and voltage, power factor and harmonic distortion rate.

[0017] Preferably, the edge computing layer consists of an industrial edge gateway, an FPGA accelerator card, and an edge AI inference engine.

[0018] Preferably, the digital twin engine constructs a digital twin of the forging hammer in the cloud based on the CAD model, material parameters, assembly relationships, and historical operating data of the forging hammer. It simulates the overall motion process and dynamic load transfer of the forging hammer based on multibody dynamics simulation, calculates the stress and strain distribution of the forging hammer components under impact load based on finite element analysis, and outputs the processed state parameters of the forging hammer at the edge computing layer. Then, the digital twin engine corrects the parameters of the digital twin of the forging hammer through an extended Kalman filter assimilation algorithm.

[0019] Preferably, the life prediction module receives the forging hammer state parameters collected by the device perception layer, the physical simulation data output by the digital twin engine, the fault diagnosis results output by the fault diagnosis model, and the historical data from the cloud database. It extracts the component degradation characteristics of the forging hammer, calculates the failure threshold of the forging hammer components based on the digital twin of the forging hammer, quantifies the degradation factors of the forging hammer components based on the Bayesian network, quantifies the degradation process of the forging hammer components based on the Wiener process degradation model, quantifies the lifespan probability distribution of the forging hammer components based on Monte Carlo simulation, and calculates the lifespan probability distribution of the forging hammer based on the lifespan probability distribution of the forging hammer components.

[0020] Preferably, after receiving the forging hammer state parameters, physical simulation data, fault diagnosis results and life probability distribution, the health assessment engine quantifies the contribution of hierarchical indicators to the health status of the forging hammer based on the analytic hierarchy process, quantifies the health status of the forging hammer based on fuzzy comprehensive evaluation, and outputs the comprehensive health score and component health score of the forging hammer based on weighted scoring, and feeds them back to the digital twin engine, fault diagnosis model and life prediction module.

[0021] Preferably, after receiving the comprehensive health score and component health score of the forging hammer, the maintenance decision optimizer quantifies the maintenance cost, production loss and equipment failure risk based on a multi-objective genetic algorithm, and outputs the forging hammer maintenance time window, spare parts demand and maintenance procedures based on mixed integer programming.

[0022] Preferably, the maintenance decision optimizer outputs a forging hammer maintenance plan to the remote maintenance decision layer, which then performs maintenance on the forging hammer through AR-assisted maintenance, remote expert collaboration, maintenance knowledge graph, and remote control and debugging.

[0023] Preferably, the device perception layer, edge computing layer, cloud operation and maintenance platform layer, and remote maintenance decision layer complete information transmission through the communication transmission layer.

[0024] Preferably, the remote operation and maintenance method is as follows:

[0025] S1: Collect the status parameters of the forging hammer during operation through the device sensing layer;

[0026] S2: The edge computing layer extracts features from the state parameters of the forging hammer through wavelet threshold denoising, trend term elimination, and normalization.

[0027] S3: The digital twin engine constructs a digital twin of the forging hammer, and the fault diagnosis model library constructs a fault diagnosis model based on the forging hammer state parameters and the digital twin of the forging hammer, and outputs the fault diagnosis results of the forging hammer;

[0028] S4: The life prediction module calculates the comprehensive health score and component health score of the forging hammer;

[0029] S5: The remote maintenance decision layer performs maintenance on the forging hammer based on the maintenance decision optimizer's output maintenance plan.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] 1. This invention designs a fault diagnosis model that uses digital twin samples of forging hammers and real samples to train together, and judges the type, location and degree of forging hammer faults. This realizes the function of accurate fault diagnosis of forging hammers, solves the problems of single fault type identification, inability to locate fault location and poor fault identification accuracy, eliminates the manual secondary judgment link, and improves the identification accuracy of forging hammer faults.

[0032] 2. This invention designs a digital twin of a forging hammer that is updated in real time according to the state parameters of the forging hammer, realizing the function of multi-physics field realistic modeling, solving the problems of disconnect between digital twin and diagnosis, low model simulation accuracy and asynchronous virtual and real, and can update the parameters of the digital twin of the forging hammer in real time according to the actual equipment status, thereby improving the accuracy of forging hammer fault prediction and life prediction.

[0033] 3. This invention is designed to extract the degradation characteristics of each component of the forging hammer to predict the service life of the forging hammer, thereby realizing the function of predicting the service life of the forging hammer. It solves the problems of poor service life prediction accuracy, unreasonable failure threshold and ignoring component dependence. It can calculate personalized failure threshold based on the information of the forging hammer components, thereby improving the accuracy and precision of service life prediction.

[0034] 4. This invention is designed to formulate a forging hammer maintenance plan based on the forging hammer health score and perform remote maintenance, realizing the function of rapid forging hammer maintenance. It solves the problems of relying on manual decision-making, inability to balance maintenance costs, production losses and equipment reliability, and low maintenance efficiency. It can accurately maintain the forging hammer and improve the maintenance efficiency of forging hammer. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the fault diagnosis process of the present invention;

[0036] Figure 2 This is a schematic diagram of the forging hammer state parameter acquisition according to the present invention;

[0037] Figure 3 This is a schematic diagram of the construction of the digital twin of the forging hammer according to the present invention;

[0038] Figure 4 This is a schematic diagram of the forging hammer service life prediction process of the present invention;

[0039] Figure 5 This is a schematic diagram of the forging hammer health scoring process of the present invention;

[0040] Figure 6 This is a schematic diagram of the forging hammer maintenance process of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Example 1: Please refer to Figure 1 , Figure 2 and Figure 3 A remote operation and maintenance system for a high-energy forging hammer includes an equipment perception layer, an edge computing layer, and a cloud operation and maintenance platform layer. The equipment perception layer is used to collect the status parameters of the forging hammer. The edge computing layer is used for real-time feature extraction and anomaly detection of the status parameters of the forging hammer. The cloud operation and maintenance platform layer consists of a digital twin engine, a fault diagnosis model library, a life prediction module, a health assessment engine, and a maintenance decision optimizer.

[0043] The digital twin engine receives the forging hammer state parameters processed by the edge computing layer and constructs a digital twin of the forging hammer. The fault diagnosis model library, based on the digital twin of the forging hammer, simulates the standard vibration spectrum, stress distribution, motion trajectory, and acoustic emission characteristics of the forging hammer fault type and fault degree. It extracts the temporal features of the forging hammer state parameters based on LSTM and the spatial features based on deep residual network. By weighting the sensor channels of the equipment perception layer, it highlights the sensor channels that are sensitive to faults. It also weights the temporal features output by LSTM to focus on the time point of the fault. Combined with the feature acquisition of stress concentration areas and high-incidence areas by the digital twin of the forging hammer, the fault diagnosis model is constructed. The fault diagnosis model outputs the probability distribution of fault type and the probability distribution of fault location, and evaluates the fault degree.

[0044] The device sensing layer consists of a vibration sensing unit, a temperature sensing unit, a pressure sensing unit, a displacement and deformation sensing unit, an acoustic sensing unit, and an energy consumption monitoring unit.

[0045] The vibration sensing unit consists of a triaxial accelerometer and a piezoelectric accelerometer, which are deployed at the hammer, anvil, guide rail and bearing to collect impact energy, rebound characteristics, guidance accuracy and bearing condition.

[0046] The temperature sensing unit consists of a PT1000 temperature sensor and an infrared thermal imager, and is deployed in the hydraulic system, motor bearings, gearbox and anvil to collect temperature distribution and temperature rise trend.

[0047] The pressure sensing unit is a high-frequency piezoresistive pressure sensor, which is deployed in the main oil circuit, accumulator and impact cylinder of the hydraulic system to collect hydraulic impact waveforms, pressure fluctuations and sealing status.

[0048] The displacement and deformation sensing unit consists of a laser displacement sensor and a resistance strain gauge, which is deployed on the guide rail to collect hammer stroke, anvil sinkage and frame deformation.

[0049] The acoustic sensing unit, consisting of a microphone array and an acoustic emission sensor, is deployed at the hammerhead for internal crack identification and cavitation detection.

[0050] The energy consumption monitoring unit consists of a smart meter and a harmonic analyzer, and is deployed in the electrical control cabinet to collect motor current and voltage, power factor and harmonic distortion rate.

[0051] The edge computing layer consists of an industrial edge gateway, an FPGA acceleration card, and an edge AI inference engine;

[0052] The device perception layer, edge computing layer, cloud operation and maintenance platform layer, and remote maintenance decision-making layer complete information transmission through the communication transmission layer;

[0053] Furthermore, based on the failure mechanism of forging hammers and industry fault statistics, the faults of each system in the forging hammer are defined and categorized into transmission system faults, execution system faults, structural system faults, hydraulic system faults, and electrical system faults. Piezoelectric accelerometers and acoustic emission sensors are deployed at the bearing housing and connecting rod to detect spindle bearing wear, excessive clearance in the connecting rod big end bearing, and crankshaft cracks. Laser displacement sensors and triaxial accelerometers are deployed at the guide rail, hammer head, and anvil to detect uneven wear of the guide rail, hammer head misalignment, and anvil sinking. Resistance strain gauges and acoustic emission sensors are deployed at the bottom of the column and the crossbeam to detect loose column foundations and fatigue cracks in the crossbeam. High-frequency piezoresistive pressure sensors and PT100 pressure sensors are deployed at the hydraulic station and impact cylinder. A temperature sensor is used to detect leaks in the main oil circuit, insufficient accumulator pressure, and failure of the impact cylinder seal. A smart meter and harmonic analyzer are deployed in the electrical control cabinet to detect three-phase imbalance in the motor and harmonic distortion in the frequency converter. The fault diagnosis model library is based on a digital twin model of the forging hammer, into which parameters corresponding to each fault mode are input, including geometric, material, and load parameters. The digital twin engine then runs to generate vibration time-domain waveforms and frequency-domain spectra, transient stress distribution cloud maps, hammer motion trajectory and velocity curves, acoustic emission signal waveforms and energy characteristics, and hydraulic pressure impact waveforms for each fault at different severity levels. Gaussian noise, electromagnetic interference, and load waves of varying intensities are added to the simulation data. The simulation of harsh environments in a real workshop ensures sufficient fault samples. Using an FPGA acceleration card (real-time edge preprocessing), the forging hammer's state parameters are filtered and denoised, anomaly detected, synchronously triggered for sampling, and preliminary feature extraction is performed before being uploaded to the cloud. Subsequently, cloud-based batch preprocessing performs multi-source data time synchronization, time-frequency conversion, spatial feature transformation, and data normalization on the forging hammer's state parameters. A two-layer bidirectional LSTM (Long Short-Term Memory) network is used to extract temporal features from the forging hammer's state parameters, capturing the long-term temporal dependencies of fault signals. A deep residual network is used to extract spatial features from the forging hammer's state parameters, extracting spatial distribution features of different frequency components from the time-frequency map and structural stress concentration features from the stress cloud map. By adaptively weighting the features of six sensor channels, the system automatically highlights the sensor channel most sensitive to the current fault, outputting weighted multi-channel time-series features. The time-step features output by the LSTM are also weighted to automatically focus on key time points where the fault occurs, suppressing irrelevant noise during steady-state operation and outputting weighted key time-series features. Furthermore, the stress concentration area coordinates and high-incidence fault location information provided by the digital twin of the forging hammer are transformed into a spatial attention weight map. Through a multi-task joint learning framework, fault type identification, fault location, and fault severity assessment branches are simultaneously output. Training is performed using a digital twin fault sample library and a small number of real fault samples, learning general fault feature patterns using these digital twin fault samples.By utilizing real-world fault samples, the fault diagnosis model rapidly adapts to the actual operating characteristics of equipment, outputting a fault diagnosis model. This model is deployed to the edge for real-time inference, and a continuous learning mechanism is established. The edge gateway receives raw data from the device's perception layer, the FPGA performs preprocessing and feature extraction, the edge AI inference engine runs the fault diagnosis model, outputs diagnostic results, and uploads them to the cloud-based operations and maintenance platform. After maintenance is completed, fault samples are labeled based on maintenance records, and the fault diagnosis model is updated based on newly added real-world fault samples.

[0054] Example 2: Please refer to Figure 1 , Figure 2 and Figure 3 A remote operation and maintenance system for a high-energy forging hammer includes an equipment perception layer, an edge computing layer, and a cloud operation and maintenance platform layer. The equipment perception layer is used to collect the status parameters of the forging hammer. The edge computing layer is used for real-time feature extraction and anomaly detection of the status parameters of the forging hammer. The cloud operation and maintenance platform layer consists of a digital twin engine, a fault diagnosis model library, a life prediction module, a health assessment engine, and a maintenance decision optimizer.

[0055] The digital twin engine receives the forging hammer state parameters processed by the edge computing layer and constructs a digital twin of the forging hammer. The fault diagnosis model library, based on the digital twin of the forging hammer, simulates the standard vibration spectrum, stress distribution, motion trajectory, and acoustic emission characteristics of the forging hammer fault type and fault degree. It extracts the temporal features of the forging hammer state parameters based on LSTM and the spatial features based on deep residual network. By weighting the sensor channels of the equipment perception layer, it highlights the sensor channels that are sensitive to faults. It also weights the temporal features output by LSTM to focus on the time point of the fault. Combined with the feature acquisition of stress concentration areas and high-incidence areas by the digital twin of the forging hammer, the fault diagnosis model is constructed. The fault diagnosis model outputs the probability distribution of fault type and the probability distribution of fault location, and evaluates the fault degree.

[0056] The edge computing layer consists of an industrial edge gateway, an FPGA accelerator card, and an edge AI inference engine.

[0057] The digital twin engine constructs a digital twin of the forging hammer in the cloud based on the CAD model, material parameters, assembly relationships, and historical operating data of the forging hammer. It simulates the overall motion process and dynamic load transfer of the forging hammer based on multibody dynamics simulation, calculates the stress and strain distribution of the forging hammer components under impact load based on finite element analysis, and after outputting the processed forging hammer state parameters at the edge computing layer, the digital twin engine corrects the parameters of the digital twin of the forging hammer through an extended Kalman filter assimilation algorithm.

[0058] The device perception layer, edge computing layer, cloud operation and maintenance platform layer, and remote maintenance decision-making layer complete information transmission through the communication transmission layer;

[0059] Furthermore, the digital twin engine constructs a digital twin of the forging hammer based on its CAD model, material parameters, assembly relationships, and historical operation records. Virtual sensors identical to those in the actual equipment are deployed in corresponding parts of the digital twin. The standard dynamic control equations for the multi-rigid-body system of the forging hammer are established based on the Lagrange equation. The pressure-flow characteristics of the hydraulic system, the contact stiffness between the hammer and the forging, and the friction characteristics of the guide rail are set. The dynamic model is preliminarily verified using factory test data to ensure that the simulated hammer speed, acceleration, and impact force do not deviate from the measured values ​​by more than 5%. The substructure method is used to divide the forging hammer system into an overall structure and key vulnerable parts. The overall structure is simulated using coarse meshes for multibody dynamics, while key vulnerable components such as connecting rods, crankshafts, columns, and hammers are analyzed using fine meshes with a mesh size of 2-5mm. The joint reactions and load time histories obtained from the multibody dynamics simulation are mapped to the corresponding nodes of the finite element model. The transient stress-strain distribution of key components during impact is solved, high-stress areas and fatigue hazard points are identified, and a multiphysics coupled simulation process of hydraulics, mechanics, and structure is established. The stress-strain distribution is calculated, and the fatigue damage parameters of the material are corrected based on the stress distribution results. The parameters requiring calibration in the digital twin model of the forging hammer are defined as state row vectors. For example, the friction coefficient, damping coefficient, and contact stiffness of each component are used. The observable features transmitted from the edge computing layer are defined as observation vectors, such as the peak hammer acceleration, hammer impact acceleration, peak impact force, peak hydraulic pressure, and peak column stress. After each hammer impact, an extended Kalman filter assimilation algorithm is executed. Based on the state estimate from the previous moment, the forging hammer digital twin is run to obtain the current state prediction value. The observed prediction value is calculated, and the actual observed value is compared with the predicted observed value to update the state estimate. The updated state estimate is then input into the corresponding parameters in the forging hammer digital twin to obtain the new state prediction value. The new observed prediction value is then calculated. The predicted values ​​are measured and compared with the actual observed values ​​to ensure that the error does not exceed 5%. After the fault diagnosis model outputs the fault location, fault type and fault severity, a mapping relationship between the fault type and the model parameters is established. Based on the fault diagnosis results, the geometric parameters, material parameters and dynamic parameters of the corresponding components in the digital twin of the forging hammer are modified. After updating the parameters, the digital twin of the forging hammer is run immediately to verify whether the observed predicted values ​​output by the updated digital twin of the forging hammer are consistent with the measured observed values. Different working condition parameters are input into the updated digital twin of the forging hammer to simulate the development rate of the predicted fault and output the predicted values ​​of the fault severity at different time points.

[0060] Example 3: Please refer to Figure 1 and Figure 4A remote operation and maintenance system for a high-energy forging hammer includes an equipment perception layer, an edge computing layer, and a cloud operation and maintenance platform layer. The equipment perception layer is used to collect the status parameters of the forging hammer. The edge computing layer is used for real-time feature extraction and anomaly detection of the status parameters of the forging hammer. The cloud operation and maintenance platform layer consists of a digital twin engine, a fault diagnosis model library, a life prediction module, a health assessment engine, and a maintenance decision optimizer.

[0061] The digital twin engine receives the forging hammer state parameters processed by the edge computing layer and constructs a digital twin of the forging hammer. The fault diagnosis model library, based on the digital twin of the forging hammer, simulates the standard vibration spectrum, stress distribution, motion trajectory, and acoustic emission characteristics of the forging hammer fault type and fault degree. It extracts the temporal features of the forging hammer state parameters based on LSTM and the spatial features based on deep residual network. By weighting the sensor channels of the equipment perception layer, it highlights the sensor channels that are sensitive to faults. It also weights the temporal features output by LSTM to focus on the time point of the fault. Combined with the feature acquisition of stress concentration areas and high-incidence areas by the digital twin of the forging hammer, the fault diagnosis model is constructed. The fault diagnosis model outputs the probability distribution of fault type and the probability distribution of fault location, and evaluates the fault degree.

[0062] The life prediction module receives the forging hammer state parameters collected by the device perception layer, the physical simulation data output by the digital twin engine, the fault diagnosis results output by the fault diagnosis model, and the historical data from the cloud database. It extracts the component degradation characteristics of the forging hammer, calculates the failure threshold of the forging hammer components based on the digital twin of the forging hammer, quantifies the degradation factors of the forging hammer components based on the Bayesian network, quantifies the degradation process of the forging hammer components based on the Wiener process degradation model, quantifies the service life probability distribution of the forging hammer components based on Monte Carlo simulation, and calculates the service life probability distribution of the forging hammer based on the service life probability distribution of the forging hammer components.

[0063] The device perception layer, edge computing layer, cloud operation and maintenance platform layer, and remote maintenance decision-making layer complete information transmission through the communication transmission layer;

[0064] Furthermore, the life prediction module receives forging hammer state parameters, physical simulation data, fault diagnosis results, and historical data. It removes outliers, missing values, and duplicates, and uses linear interpolation to complete a small number of missing data points. For typical failure modes of different components of the forging hammer, it extracts corresponding multi-dimensional degradation features. It uses the root mean square value of vibration, kurtosis, envelope spectrum peak value, and acoustic emission energy to determine whether the spindle bearing and connecting rod bearing have experienced fatigue wear or pitting. It uses the lateral vibration amplitude, hammer head offset, and friction coefficient to determine whether the guide rail has experienced uneven wear or increased clearance. It uses the stress amplitude, strain change rate, and acoustic emission event rate to determine whether the connecting rod or crankshaft has experienced fatigue crack propagation. Finally, it uses the pressure fluctuation amplitude, oil temperature rise rate, and internal leakage to determine the hydraulic cylinder seal. To determine if wear and leakage have occurred, the vibration amplitude of the foundation, strain of the column, and settlement of the anvil are used to assess whether the column or anvil has experienced fatigue deformation or loosening of the foundation anchor. Principal component analysis is used to reduce the dimensionality of the multi-dimensional features of each component, extracting the first three principal components. These principal components are then transformed into a single dimensionless degradation index using a weighted summation method. This degradation index is further normalized to a value range of 0-1, forming a continuous degradation trend sequence. A digital twin engine is used to calculate the ultimate stress distribution of the forging hammer components under the rated maximum impact energy. Based on the material's SN curve and Miner's linear cumulative damage theory, the theoretical fatigue life of each component under ideal working conditions is calculated. Based on the correspondence between the theoretical fatigue life and the degradation index, the degradation trend of the component is determined. The initial failure threshold of the component is determined and corrected based on the current fault severity output by the fault diagnosis model. Further corrections are made based on the service life of the forging hammer, historical maintenance records, and actual operating conditions. A three-layer Bayesian network is used to model the degradation rate of each component of the forging hammer. Factors such as impact energy, impact frequency, ambient temperature, lubrication status, and fault severity constitute the parent node layer, the intermediate node layer represents the average degradation rate of the component, and the child node layer represents the degradation index of each component. Degradation simulation data under different operating conditions is generated using a digital twin of the forging hammer. The simulation data is statistically analyzed to obtain the prior probability distribution of the degradation rate. Combined with historical data and expert experience, the prior probability distribution of each parent node is determined. The system employs a Bayesian probability distribution method. After the equipment's sensing layer acquires new forging hammer state parameters, it updates the posterior probability distribution of the degradation rate using a Bayesian formula. The degradation index sequence and the posterior probability distribution of the degradation rate are input into the Wiener process degradation model, which outputs the probability density function of the component's remaining service life. Based on the Wiener process degradation model and the posterior probability distribution of the degradation rate, a Monte Carlo simulation is performed, outputting a point estimate and confidence interval for the component's remaining service life. A Bayesian network is used to model the failure dependencies between components, such as guide rail wear leading to hammer head misalignment, which accelerates connecting rod bearing degradation. Based on the predicted remaining service life of each component and the dependencies between components, the system outputs a point estimate and confidence interval for the overall remaining service life of the forging hammer system.The communication transmission layer inputs new forging hammer condition parameters, fault diagnosis results, and maintenance records into the life prediction model to update the remaining service life of the forging hammer system's components and the overall remaining service life.

[0065] Example 4: Please refer to Figure 1 and Figure 5 A remote operation and maintenance system for a high-energy forging hammer includes an equipment perception layer, an edge computing layer, and a cloud operation and maintenance platform layer. The equipment perception layer is used to collect the status parameters of the forging hammer. The edge computing layer is used for real-time feature extraction and anomaly detection of the status parameters of the forging hammer. The cloud operation and maintenance platform layer consists of a digital twin engine, a fault diagnosis model library, a life prediction module, a health assessment engine, and a maintenance decision optimizer.

[0066] The digital twin engine receives the forging hammer state parameters processed by the edge computing layer and constructs a digital twin of the forging hammer. The fault diagnosis model library, based on the digital twin of the forging hammer, simulates the standard vibration spectrum, stress distribution, motion trajectory, and acoustic emission characteristics of the forging hammer fault type and fault degree. It extracts the temporal features of the forging hammer state parameters based on LSTM and the spatial features based on deep residual network. By weighting the sensor channels of the equipment perception layer, it highlights the sensor channels that are sensitive to faults. It also weights the temporal features output by LSTM to focus on the time point of the fault. Combined with the feature acquisition of stress concentration areas and high-incidence areas by the digital twin of the forging hammer, the fault diagnosis model is constructed. The fault diagnosis model outputs the probability distribution of fault type and the probability distribution of fault location, and evaluates the fault degree.

[0067] After receiving the forging hammer state parameters, physical simulation data, fault diagnosis results and life probability distribution, the health assessment engine quantifies the contribution of hierarchical indicators to the health status of the forging hammer based on the analytic hierarchy process, quantifies the health status of the forging hammer based on fuzzy comprehensive evaluation, and outputs the comprehensive health score and component health score of the forging hammer based on weighted scoring, and feeds them back to the digital twin engine, fault diagnosis model and life prediction module.

[0068] The maintenance decision optimizer receives the comprehensive health score and component health score of the forging hammer, and quantifies the maintenance cost, production loss and equipment failure risk based on a multi-objective genetic algorithm. It then outputs the forging hammer maintenance time window, spare parts requirements and maintenance procedures based on mixed integer programming.

[0069] The maintenance decision optimizer outputs a forging hammer maintenance plan to the remote maintenance decision layer, which then maintains the forging hammer through AR-assisted maintenance, remote expert collaboration, maintenance knowledge graph, and remote control and debugging.

[0070] The device perception layer, edge computing layer, cloud operation and maintenance platform layer, and remote maintenance decision-making layer complete information transmission through the communication transmission layer;

[0071] Furthermore, a three-level evaluation index system was established based on the physical structure and functional logic of the forging hammer. The first-level index is the comprehensive health score of the forging hammer; the second-level indexes are the health scores of the transmission system, execution system, structural system, hydraulic system, and electrical system; and the third-level index is the health score of the core components of the forging hammer. A judgment matrix was constructed using the 1-9 scale method, and the initial weights of each level of indexes were calculated and consistency checks were performed. For example, the weight of the transmission system is 0.35, the weight of the execution system is 0.25, the weight of the hydraulic system is 0.2, the weight of the structural system is 0.15, the weight of the electrical system is 0.05, the weight of the main shaft bearing is 0.59, the weight of the crankshaft is 0.31, and the weight of the connecting rod is 0.1. When the severity of a component's failure exceeds 80 points, its weight automatically increases by 50%. When the remaining service life of a component is less than 30 days, its weight automatically increases by 30%. When the health score of a component drops by more than 10% for 7 consecutive days, its weight automatically increases by 20%. The raw data of each tertiary indicator is converted into a health score of 0-100, with 90-100 being excellent, 75-89 being good, 60-74 being average, 40-59 being poor, and 0-39 being extremely poor. Numerical indicators such as temperature, vibration amplitude, and pressure use a trapezoidal membership function. The failure severity indicator is directly mapped to the health score = 100 - failure severity. The health score corresponding to the lifespan index is calculated as (remaining service life / design service life) × 100. The membership degree of each index to its respective health level is calculated, and the membership degree of each index to its five health levels is calculated. A fuzzy evaluation matrix is ​​constructed, and a weighted average method is used to calculate the health score of each tertiary index. Based on the weights of the tertiary indexes, components, and subsystems, the health scores of components, subsystems, and the forging hammer as a whole are calculated with weights at each level. Corresponding response strategies are triggered according to the health level: Excellent corresponds to normal operation and daily automatic inspection; Good corresponds to normal operation and intensive inspection every 3 days; Average corresponds to triggering a preventative maintenance strategy; and Poor corresponds to triggering an emergency maintenance strategy. The extreme difference immediately triggers a remote shutdown command. After receiving the health score of the forging hammer, the maintenance decision optimizer automatically determines the maintenance content. When the health score is greater than 80, routine maintenance such as automatic lubrication, parameter calibration, and fastener inspection is performed. When the health score is between 60 and 80, preventive maintenance such as replacement of vulnerable parts, gap adjustment, and precision calibration is performed. When the health score is between 40 and 60, restorative maintenance such as disassembly and replacement of faulty parts and performance restoration is performed. When the health score is less than 40, major overhaul operations such as complete machine disassembly, comprehensive inspection, and batch replacement of key components are performed. A minimization objective function is constructed based on the total maintenance cost, total production loss, and equipment failure risk. A multi-objective genetic algorithm generates the optimal maintenance time window scheme. Within the selected time window, mixed integer programming is used to confirm spare parts requirements and maintenance procedures. For parameter-based faults that do not require disassembly, maintenance personnel can complete the task remotely via control and debugging. When on-site maintenance is required, personnel can wear AR glasses and complete the maintenance with the assistance of digital twins and knowledge graphs. For complex large-scale maintenance, on-site personnel can operate the equipment in conjunction with remote expert guidance. After maintenance is completed, the forging hammer parameters are re-collected through the equipment perception layer, and the parameters of the forging hammer digital twin, fault diagnosis model, life prediction model, and maintenance decision optimizer are updated.

[0072] Example 5: Please refer to Figure 2 and Figure 3A remote operation and maintenance system for a high-energy forging hammer, wherein the equipment sensing layer consists of a vibration sensing unit, a temperature sensing unit, a pressure sensing unit, a displacement and deformation sensing unit, an acoustic sensing unit, and an energy consumption monitoring unit.

[0073] The vibration sensing unit consists of a triaxial accelerometer and a piezoelectric accelerometer, which are deployed at the hammer, anvil, guide rail and bearing to collect impact energy, rebound characteristics, guidance accuracy and bearing condition.

[0074] The temperature sensing unit consists of a PT1000 temperature sensor and an infrared thermal imager, and is deployed in the hydraulic system, motor bearings, gearbox and anvil to collect temperature distribution and temperature rise trend.

[0075] The pressure sensing unit is a high-frequency piezoresistive pressure sensor, which is deployed in the main oil circuit, accumulator and impact cylinder of the hydraulic system to collect hydraulic impact waveforms, pressure fluctuations and sealing status.

[0076] The displacement and deformation sensing unit consists of a laser displacement sensor and a resistance strain gauge, which is deployed on the guide rail to collect hammer stroke, anvil sinkage and frame deformation.

[0077] The acoustic sensing unit, consisting of a microphone array and an acoustic emission sensor, is deployed at the hammerhead for internal crack identification and cavitation detection.

[0078] The energy consumption monitoring unit consists of a smart meter and a harmonic analyzer, and is deployed in the electrical control cabinet to collect motor current and voltage, power factor and harmonic distortion rate.

[0079] The edge computing layer consists of an industrial edge gateway, an FPGA acceleration card, and an edge AI inference engine;

[0080] The digital twin engine constructs a digital twin of the forging hammer in the cloud based on the CAD model, material parameters, assembly relationships, and historical operating data of the forging hammer. It simulates the overall motion process and dynamic load transfer of the forging hammer based on multibody dynamics simulation, calculates the stress and strain distribution of the forging hammer components under impact load based on finite element analysis, and after outputting the processed forging hammer state parameters at the edge computing layer, the digital twin engine corrects the parameters of the digital twin of the forging hammer through an extended Kalman filter assimilation algorithm.

[0081] Furthermore, during the forging process, the impact energy and rebound characteristics of the hammer are collected by a vibration sensing unit, the hydraulic impact waveform and impact cylinder pressure are collected by a pressure sensing unit, the hammer stroke and impact speed are collected by a displacement sensing unit, and the stress distribution of the anvil and the deformation resistance of the forging are collected by a strain gauge. Utilizing the parallel computing capabilities of the FPGA accelerator card, the raw process data is preprocessed to extract key process features such as peak impact force, impact energy, forming time, and rebound amount. After receiving information transmitted from the edge computing layer, the digital twin engine modifies the parameters and simulates the digital twin of the forging hammer, simulating the forging process, calculating the stress and strain distribution inside the forging, predicting the dimensional accuracy and internal defects of the forging, and assessing the wear state of the die. Subsequently, the simulation results are compared with the standard process to identify process deviations, generate process parameter adjustment suggestions, such as adjustments to impact energy, number of impacts, and hydraulic pressure, and control the forging hammer to make corresponding adjustments to ensure that the forging quality is guaranteed and the forging hammer damage is reduced during the forging process.

[0082] Working principle: By deploying a network of six types of sensors in the forging hammer body and its surrounding environment, vibration, temperature, pressure, displacement, acoustic and energy consumption data are collected in all directions. The edge computing layer uses an industrial edge gateway to aggregate data and uses an FPGA acceleration card to complete filtering and noise reduction, anomaly detection and synchronous trigger sampling. Only 100ms of transient data before and after the forging hammer strike is transmitted, compressing the original data volume by more than 90%. The edge AI inference engine realizes real-time inference of lightweight models. The communication transmission layer adopts the PTP precise time protocol to ensure microsecond-level synchronization of multi-source data, providing a high-quality and low-latency data foundation for cloud analysis.

[0083] The digital twin engine constructs an initial 3D model based on CAD models, material parameters, and assembly relationships. It achieves multi-physics coupled simulation through multibody dynamics and finite element analysis. After each impact, the extended Kalman filter algorithm is used to calibrate key parameters such as friction, damping, and stiffness using measured data, keeping the simulation error within 5%. The engine is pre-trained using fault samples generated by the digital twin of the forging hammer, and then fine-tuned using a small number of real samples. It integrates LSTM temporal features, deep residual network spatial features, and multi-scale attention mechanisms, and outputs the fault type, location, and severity of the forging hammer.

[0084] The health assessment engine establishes a three-level assessment index system of whole machine, subsystem, and component. It uses the analytic hierarchy process to calculate the initial weights and dynamically adjusts them according to the fault status. It transforms multi-source data into a health score of 0-100 through fuzzy comprehensive evaluation. The life prediction module integrates physical simulation and real-time status data, quantifies degradation factors based on Bayesian networks, models degradation trends through Wiener processes, and uses Monte Carlo simulation to propagate uncertainty. It outputs the remaining service life and confidence interval of each component and the whole machine.

[0085] The maintenance decision optimizer aims to minimize maintenance costs, production losses, and equipment risks. It uses a multi-objective genetic algorithm to generate the optimal maintenance time window, optimizes spare parts demand and maintenance procedures through mixed integer programming, automatically matches the maintenance mode to the forging hammer based on the hammer's health level, and updates the hammer's status parameters after maintenance.

[0086] 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 invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A remote operation and maintenance system for a high-energy forging hammer, characterized in that: It includes an equipment perception layer, an edge computing layer, and a cloud-based operation and maintenance platform layer. The equipment perception layer is used to collect the status parameters of the forging hammer. The edge computing layer is used for real-time feature extraction and anomaly detection of the forging hammer status parameters. The cloud-based operation and maintenance platform layer consists of a digital twin engine, a fault diagnosis model library, a life prediction module, a health assessment engine, and a maintenance decision optimizer. The digital twin engine receives the forging hammer state parameters processed by the edge computing layer and constructs a digital twin of the forging hammer. The fault diagnosis model library, based on the forging hammer digital twin, simulates the standard vibration spectrum, stress distribution, motion trajectory, and acoustic emission characteristics of the forging hammer fault type and fault degree. It extracts the temporal features of the forging hammer state parameters based on LSTM and the spatial features based on deep residual networks. By weighting the sensor channels of the equipment perception layer, it highlights the sensor channels sensitive to faults. It also weights the temporal features output by LSTM to focus on the time point of fault occurrence. Combining the feature acquisition of stress concentration areas and high-incidence areas by the forging hammer digital twin, the fault diagnosis model is constructed. The fault diagnosis model outputs the probability distribution of fault type and fault location and evaluates the fault degree.

2. The remote operation and maintenance system for a high-energy forging hammer according to claim 1, characterized in that: The device sensing layer consists of a vibration sensing unit, a temperature sensing unit, a pressure sensing unit, a displacement and deformation sensing unit, an acoustic sensing unit, and an energy consumption monitoring unit. The vibration sensing unit consists of a triaxial accelerometer and a piezoelectric accelerometer, which are deployed at the hammer, anvil, guide rail and bearing to collect impact energy, rebound characteristics, guidance accuracy and bearing condition. The temperature sensing unit consists of a PT1000 temperature sensor and an infrared thermal imager, and is deployed in the hydraulic system, motor bearings, gearbox and anvil to collect temperature distribution and temperature rise trend. The pressure sensing unit is a high-frequency piezoresistive pressure sensor, which is deployed in the main oil circuit, accumulator and impact cylinder of the hydraulic system to collect hydraulic impact waveforms, pressure fluctuations and sealing status. The displacement and deformation sensing unit consists of a laser displacement sensor and a resistance strain gauge, which is deployed on the guide rail to collect hammer stroke, anvil sinkage and frame deformation. The acoustic sensing unit, consisting of a microphone array and an acoustic emission sensor, is deployed at the hammerhead for internal crack identification and cavitation detection. The energy consumption monitoring unit consists of a smart meter and a harmonic analyzer, and is deployed in the electrical control cabinet to collect motor current and voltage, power factor and harmonic distortion rate.

3. The remote operation and maintenance system for a high-energy forging hammer according to claim 1, characterized in that: The edge computing layer consists of an industrial edge gateway, an FPGA accelerator card, and an edge AI inference engine.

4. The remote operation and maintenance system for a high-energy forging hammer according to claim 1, characterized in that: The digital twin engine constructs a digital twin of the forging hammer in the cloud based on the CAD model, material parameters, assembly relationships, and historical operating data of the forging hammer. It simulates the overall motion process and dynamic load transfer of the forging hammer based on multibody dynamics simulation, calculates the stress and strain distribution of the forging hammer components under impact load based on finite element analysis, and outputs the processed state parameters of the forging hammer at the edge computing layer. The digital twin engine then corrects the parameters of the digital twin of the forging hammer through an extended Kalman filter assimilation algorithm.

5. The remote operation and maintenance system for a high-energy forging hammer according to claim 4, characterized in that: The life prediction module receives the forging hammer state parameters collected by the device perception layer, the physical simulation data output by the digital twin engine, the fault diagnosis results output by the fault diagnosis model, and the historical data from the cloud database. It extracts the component degradation characteristics of the forging hammer, calculates the failure threshold of the forging hammer components based on the digital twin of the forging hammer, quantifies the degradation factors of the forging hammer components based on the Bayesian network, quantifies the degradation process of the forging hammer components based on the Wiener process degradation model, quantifies the service life probability distribution of the forging hammer components based on Monte Carlo simulation, and calculates the service life probability distribution of the forging hammer based on the service life probability distribution of the forging hammer components.

6. The remote operation and maintenance system for a high-energy forging hammer according to claim 5, characterized in that: The health assessment engine receives the forging hammer's state parameters, physical simulation data, fault diagnosis results, and life probability distribution. It then quantifies the contribution of hierarchical indicators to the forging hammer's health status based on the analytic hierarchy process (AHP), quantifies the forging hammer's health status based on fuzzy comprehensive evaluation, and outputs the forging hammer's comprehensive health score and component health score based on weighted scoring. These scores are then fed back to the digital twin engine, fault diagnosis model, and life prediction module.

7. The remote operation and maintenance system for a high-energy forging hammer according to claim 6, characterized in that: After receiving the comprehensive health score and component health score of the forging hammer, the maintenance decision optimizer quantifies maintenance costs, production losses and equipment failure risks based on a multi-objective genetic algorithm, and outputs the forging hammer maintenance time window, spare parts requirements and maintenance procedures based on mixed integer programming.

8. The remote operation and maintenance system for a high-energy forging hammer according to claim 7, characterized in that: The maintenance decision optimizer outputs a forging hammer maintenance plan to the remote maintenance decision layer, which then performs maintenance on the forging hammer through AR-assisted maintenance, remote expert collaboration, maintenance knowledge graph, and remote control and debugging.

9. The remote operation and maintenance system for a high-energy forging hammer according to claim 8, characterized in that: The device perception layer, edge computing layer, cloud operation and maintenance platform layer, and remote maintenance decision-making layer complete information transmission through the communication transmission layer.

10. A method for remote operation and maintenance of a high-energy forging hammer, applicable to the remote operation and maintenance system for a high-energy forging hammer as described in any one of claims 1-9, characterized in that: The remote operation and maintenance method is as follows: S1: Collect the status parameters of the forging hammer during operation through the device sensing layer; S2: The edge computing layer extracts features from the state parameters of the forging hammer through wavelet threshold denoising, trend term elimination, and normalization. S3: The digital twin engine constructs a digital twin of the forging hammer, and the fault diagnosis model library constructs a fault diagnosis model based on the forging hammer state parameters and the digital twin of the forging hammer, and outputs the fault diagnosis results of the forging hammer; S4: The life prediction module calculates the comprehensive health score and component health score of the forging hammer; S5: The remote maintenance decision layer performs maintenance on the forging hammer based on the maintenance decision optimizer's output maintenance plan.

Images