Electrical fault positioning system and method for electric anastomat

Abstract

The invention discloses an electrical fault positioning system and method for an electric anastomat, and belongs to the technical field of medical equipment fault detection.The electrical, thermal and mechanical operation data of equipment are comprehensively collected through a multi-modal sensing subsystem, and real-time feature extraction and AI reasoning are conducted through an edge calculation unit; according to the method, a deep learning model is adopted to carry out preliminary fault classification, a strategy of combining a confidence coefficient mechanism and digital twinborn simulation verification is introduced, and multi-physics field coupling simulation and root cause analysis are carried out on a low confidence coefficient result, so that accurate positioning of faults is realized. According to the method, physical simulation and data-driven intelligence are fused, the problems that complex fault diagnosis is high in uncertainty and difficult to position are effectively solved, the accuracy, reliability and operation and maintenance efficiency of fault diagnosis of the electric anastomat are remarkably improved, and predictive maintenance capacity is achieved.

Description

Technical Field

[0001] This invention relates to the field of electric stapler detection technology, specifically to an electrical fault location system and method for electric staplers. Background Technology

[0002] As a key piece of equipment in modern surgery, the reliability and accuracy of electric staplers directly affect the success of the operation and the safety of the patient. With the development of technology, electric staplers integrate complex motor drives, precision control circuits and power electronic modules. Under long-term high-load operation, their electrical systems are prone to various electrical faults due to insulation aging, component failure, poor contact, sudden load changes, etc., such as winding short circuits, power transistor open circuits, grounding faults, and signal abnormalities.

[0003] Currently, fault diagnosis and localization of electric staplers mainly rely on the following methods: Manual troubleshooting: Repair personnel rely on tools such as multimeters and oscilloscopes to perform manual measurements, combining fault symptoms with their personal experience to make a judgment. This method is inefficient, heavily dependent on expert knowledge, and difficult to accurately locate intermittent or complex faults.

[0004] Basic status monitoring: Some systems integrate simple voltage or temperature alarm functions, but the monitoring dimensions are limited. They can usually only issue alarms after a fault occurs or when the situation deteriorates significantly, and cannot achieve early warning and accurate root cause location.

[0005] Traditional threshold diagnosis uses a fixed threshold to judge a single parameter. This method lacks comprehensive analysis of weak early-stage fault characteristics and multi-parameter coupling relationships, resulting in high false alarm and false negative rates, and it cannot cope with complex and ever-changing surgical conditions.

[0006] Furthermore, existing fault diagnosis methods generally suffer from the following technical bottlenecks: a lack of simultaneous perception and fusion analysis of multiple physical quantities such as electrical, thermal, and mechanical components, making it difficult to comprehensively capture the full picture of the fault; diagnostic methods based on simple rules or single models have poor ability to distinguish fault modes with similar characteristics, resulting in low confidence in diagnostic results; and most methods can only determine whether a fault exists, but cannot accurately locate specific electrical phases, physical coordinates, or components, making rapid repair difficult. Therefore, there is an urgent need in this field for an electrical fault location system and method that can achieve accurate, rapid, and automated operation to improve the intelligent level of operation and maintenance of electric staplers and enhance surgical safety. Summary of the Invention

[0007] To address the above problems, this invention proposes an electrical fault location system for an electric stapler, comprising: The multimodal sensing subsystem is deployed on the drive motor, power supply module, control circuit and key mechanical components of the electric anastomosis device to collect multi-dimensional operating data in terms of electricity, heat and mechanics. The edge computing unit is communicatively connected to the multimodal sensing subsystem and has a real-time sampling engine, a signal preprocessing module, and an AI inference acceleration module, which are used to synchronously sample, extract features, and identify fault modes of the collected data. The digital twin model, built in the cloud or locally, is synchronized with the edge computing unit and is used to perform fault simulation and verification when the fault identification confidence level is lower than a set threshold. A fault diagnosis engine, integrated into the edge computing unit and / or the cloud, is used for fault classification and localization based on multi-source feature fusion and deep learning models. The human-machine interaction module is used to display fault information, receive user commands, and support remote maintenance operations.

[0008] Furthermore, the multimodal sensing subsystem includes a current sensor, a temperature sensor, an acceleration sensor, and an optical position sensor, which are respectively deployed on the motor stator, motor bearing, power supply module, control circuit, and overall body, and are connected to the edge computing unit through the Ethernet-TSN bus, with a timing synchronization error of no more than 1µs.

[0009] Furthermore, the edge computing unit uses a RISC-V AI accelerator as its core processor, is equipped with a real-time sampling engine to achieve multi-channel synchronous sampling, a signal preprocessing module to complete FIR filtering and FFT operations, and an AI inference module to execute a CNN-LSTM dual-stream network model, with a single inference time of less than 5ms.

[0010] Furthermore, the digital twin model is an electromagnetic-thermal-mechanical multiphysics coupled model constructed based on the finite element method. It supports the injection of fault simulation library. When the confidence level output by the fault diagnosis engine is lower than the threshold, simulation verification is automatically triggered. The fault type and location are confirmed by comparing the correlation coefficient between the simulation waveform and the actual waveform.

[0011] Furthermore, the electrical fault location system also includes a predictive maintenance module, which predicts the remaining lifespan based on a hybrid model of random forest and ResNet, and triggers an early warning when the health index HI is lower than the index threshold.

[0012] This invention also proposes a method for locating electrical faults in an electric stapler, implemented through the aforementioned system, comprising the following steps: S1. By deploying multimodal sensors at key parts of the electric stapler, multi-dimensional operational data of electricity, heat, and mechanics are collected synchronously, and the collected raw data is preprocessed by noise reduction and normalization. S2. Extract time-domain, frequency-domain, and kernel density-based statistical features from the preprocessed data to form a multi-source feature vector; S3. Input the multi-source feature vector into a pre-trained deep learning model for fault classification and output its confidence level; when the confidence level is lower than a preset threshold, trigger the digital twin simulation verification process. S4. The current operating conditions and the multi-source feature vectors are used as boundary conditions and injected into a high-fidelity digital twin model for multi-physics coupling simulation. The simulation data and the measured data are compared to confirm the fault mode and locate the root cause. S5. Based on the confirmed fault mode, output fault location information including electrical phase, mechanical position or thermal coordinates, and generate a maintenance guidance report.

[0013] Furthermore, step S1 specifically includes: The three-phase electrical parameters of the drive motor are collected using current sensors; Mechanical vibration signals are collected using vibration sensors; Temperature sensors are used to monitor the thermal state of key components; Optical position sensors are used to acquire displacement information of the actuator; Data acquisition from all sensors is synchronized by a master clock.

[0014] Furthermore, step S2 specifically includes: Calculate the effective value, peak value, peak-to-peak value, waveform factor, impulse factor, rise / fall time, and zero-crossing distortion of the current and vibration signals; Windowed interpolation FFT analysis is performed on the current signal to extract the amplitude and phase of the fundamental wave and each harmonic, and the total harmonic distortion rate is calculated. Spectral analysis of the vibration signal is performed to extract the bearing's characteristic frequencies and sideband components.

[0015] Furthermore, following step S5, a predictive maintenance step is also included: Based on cumulative operating time, load cycles, temperature rise rate, and vibration spectrum drift characteristics, a hybrid model combining random forest and residual network is used to predict the remaining service life and health index of the equipment; when the health index is lower than the index threshold, the system issues an early warning.

[0016] Furthermore, in step S4, the high-fidelity digital twin model is an electromagnetic-thermal-mechanical multiphysics coupling model constructed based on the finite element method; The comparison between the simulation data and the measured data is achieved by calculating the cross-correlation coefficient between the simulation waveform and the field sampled waveform. When the correlation coefficient is greater than 0.92, the fault mode matching is confirmed.

[0017] Compared with the prior art, the present invention has the following beneficial effects: It achieves comprehensive and high-precision multimodal perception and data fusion: by deploying a multimodal sensor network covering electrical, thermal and mechanical dimensions, and using high-precision time synchronization technology, a comprehensive digital image of the operating status of the electric stapler is constructed, providing a rich and reliable data foundation for fault diagnosis and fundamentally overcoming the limitations of single sensor monitoring.

[0018] Employing deep learning models such as CNN-LSTM for multi-source feature fusion and fault mode recognition can capture complex nonlinear features and temporal dependencies, resulting in high diagnostic accuracy. In particular, the introduction of a confidence-based uncertainty management mechanism automatically triggers digital twin simulation verification for cases with model uncertainty, combining it with the physical model for secondary diagnosis, significantly reducing the risk of misdiagnosis and missed diagnosis.

[0019] The high-fidelity multiphysics coupling model built by digital twin technology can reproduce the fault scenario in virtual space. It can not only identify the fault type, but also accurately locate the specific electrical phase, mechanical position and even the component level, and intuitively display the hot spots, magnetic field distortion and other phenomena at the fault point, providing highly targeted guidance for maintenance.

[0020] The system integrates end-to-end capabilities from real-time diagnosis, simulation verification, remote maintenance to predictive maintenance. When an unknown new type of fault is detected, it can automatically initiate expert system intervention and model incremental learning processes, enabling the system to continuously learn and self-evolve, and constantly expand its diagnostic knowledge base.

[0021] Through millisecond-level edge AI inference, automated fault report generation, visualized maintenance guidance, and OTA remote upgrades, troubleshooting and repair time is significantly reduced. Meanwhile, the predictive maintenance function based on the Health Index (HI) can detect potential faults in advance and issue warnings, transforming passive maintenance into proactive maintenance, effectively extending equipment lifespan and ensuring the continuity and safety of surgeries. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the structure of the multimodal sensing subsystem of the present invention; Figure 2 This is a schematic diagram of the fault mode matching process of the present invention; Figure 3 This is a schematic diagram of the fault location output process of the present invention. Detailed Implementation

[0023] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0024] Example 1 Electrical fault location system for electric staplers 1. Hardware layer 1.1 Multimodal Sensing Subsystem To achieve electrical health monitoring of the electric stapler throughout its entire life cycle, multiple sensors are deployed in key areas to form a sensing network covering the three dimensions of electricity, heat, and mechanics.

[0025] Specifically, various sensors, such as voltage / current sensors, temperature sensors, vibration / acceleration sensors, and optical position sensors, are installed on the drive motor, power supply module, control circuit, and key mechanical components of the electric anastomosis device to achieve comprehensive perception of electrical parameters, thermal state, and motion state.

[0026] In a preferred embodiment, the sensor placement locations and types are shown in Table 1, and the structural schematic diagram is shown below. Figure 1 As shown.

[0027] Table 1

[0028] All sensors are connected to the edge computing unit via CAN-FD or Ethernet-TSN bus to ensure that the timing synchronization error does not exceed 1µs and to meet the phase alignment requirements of high-speed current waveforms.

[0029] 1.2 Edge Computing Unit It uses an Arm Cortex-M55 + NPU or an equivalent RISC-V AI accelerator as its core processor, and is equipped with 2GB DDR4 and 8GB eMMC storage. The hardware functionalities are divided as follows: Real-time sampling engine: Based on DMA double buffering, it achieves multi-channel synchronous sampling above 10kHz. The sampled data is directly written to the high-speed cache to avoid jitter caused by CPU intervention.

[0030] Signal preprocessing module: Performs FIR filtering, baseline drift removal, FFT and other operations at the hardware level, and outputs time-domain and frequency-domain feature vectors.

[0031] AI inference acceleration: The NPU adopts an 8-bit quantization model, and the time for single multi-sensor fusion feature classification is less than 5ms, meeting the "millisecond-level" response requirements of the operating room.

[0032] Safety self-test circuit: After power-on, it automatically detects power isolation, sensor connectivity, and communication link integrity. If the test fails, it immediately enters a safe shutdown state and reports the failure.

[0033] Redundancy design: Key modules adopt dual-mode redundancy and automatic fault switching is achieved through a watchdog timer.

[0034] 1.3 Safety Isolation and Redundant Power Supply Dual-channel isolated power supply: Each channel uses a 230V→24V isolated converter, and the output is connected to the control circuit through optocoupler isolation to prevent high-voltage side faults from spreading to the low-voltage side.

[0035] Self-test circuit: Performs a complete test on power supply voltage, output ripple, and grounding resistance every 30 seconds, and writes the test results to the safety log in real time.

[0036] Arc detection: A photoelectric arc sensor is connected in parallel on the high-voltage side, with a threshold set to 10kV·µs. When an arc is detected, an emergency cut-off and audible and visual alarm are triggered.

[0037] 1.4 Communication Module 5G / Wi-Fi Dual Mode: Supports Sub-6GHz 5G NR and 802.11ax dual-band, automatically switching to ensure network availability.

[0038] TLS 1.3 and Chinese national cryptographic standards SM2 / SM4: end-to-end encryption, using hardware security modules (HSM) for key management to prevent man-in-the-middle attacks.

[0039] Two-way OTA channel: Supports differential block upgrades, with a single upgrade data volume of ≤2MB. The signature is verified throughout the upgrade process to ensure firmware integrity.

[0040] 2. Data Layer 2.1 Real-time data stream Sampling frequencies: current / voltage 10kHz, temperature 1Hz, vibration 5kHz, displacement 2kHz.

[0041] Local compression: The original waveform is compressed using Lossless LZ4 with a compression ratio of approximately 2.5:1 and a compression / decompression time of <1ms.

[0042] Secure channel: Integrity verification of each data frame is performed using TLS and MAC to prevent tampering.

[0043] 2.2 Cloud Data Lake and Historical Database Object storage: uses tiered cold and hot storage to meet regulatory requirements for data retention.

[0044] Metadata index: Each waveform corresponds to a unique UUID and records metadata such as device SN, sampling time, sensor location, and anomaly tag for easy retrieval later.

[0045] Big Data Platform: Based on Apache Spark and Delta Lake, it achieves unified batch / stream processing and provides feature engineering support for model training.

[0046] 3. Software layer 3.1 Digital Twin Model Geometric modeling: Export STL / STEP files using CATIA / SolidWorks, and combine them with ANSYS Maxwell and Fluent to complete the electromagnetic field and thermal field coupling simulation.

[0047] Real-time synchronization: The edge unit pushes the key status of current, temperature and displacement to the cloud in JSON format via MQTT. The cloud digital twin engine corrects the model status through Kalman Filter to achieve real-time synchronization.

[0048] Fault simulation library: More than 20 typical fault scenarios, such as short circuit, open circuit, grounding, overheating, and drive failure, are pre-built on the digital twin, generating corresponding waveform, temperature, and vibration feature libraries for subsequent model comparison.

[0049] 3.2 Fault Diagnosis Engine 3.2.1 Signal Analysis Fourier Period Algorithm: Perform Short Time Fourier Transform (STFT) on the current / voltage waveform to extract the fundamental amplitude, harmonic sequence, and instantaneous phase.

[0050] Kernel density estimation (KDE): Probability density modeling of harmonic distortion (THD) for each sampling window, calculation of anomaly probability, and rapid location of instantaneous short circuits or grounding.

[0051] 3.2.2 Machine Learning Feature fusion: Time-domain features such as mean, peak value, and rise time, frequency-domain features such as THD and harmonic phase, and statistical features of temperature / vibration are concatenated into a 128-dimensional vector.

[0052] Model structure: A CNN-LSTM dual-stream network is used. The CNN is responsible for extracting local time-frequency patterns, and the LSTM captures the temporal dependencies across windows.

[0053] Training strategy: Transfer learning was used, with pre-training on a publicly available motor fault dataset followed by fine-tuning on 100,000 real surgical data collected by our system. Model accuracy ≥ 96%, recall ≥ 94%.

[0054] 3.2.3 Digital Twin Simulation Verification When the model confidence level is less than 0.85, the system injects the current feature vector into the digital twin to reproduce the same load, speed, and power supply conditions, and runs electromagnetic-thermal-structural multiphysics coupling simulation.

[0055] The waveform obtained from the simulation is cross-correlated with the waveform sampled on site. If the correlation coefficient is >0.92, the fault type is confirmed and the fault location is determined; otherwise, it is marked as an unknown / novel fault and enters the manual expert review process.

[0056] 3.3 Remote Maintenance and OTA Fault confirmation: The platform automatically generates a fault report, including fault code, location coordinates, possible causes, and suggested handling methods, and pushes it to the mobile device of the maintenance personnel; Firmware upgrade: For software defects, the system generates differential patches and automatically distributes them via OTA with secure signature. The entire upgrade process is logged. Hardware replacement guide: If the problem is identified as a hardware failure, the system provides the part number, disassembly and assembly video, and spare parts list to help with quick on-site replacement.

[0057] 3.4 Predictive Maintenance Feature library: Features such as cumulative running time, cumulative load cycles, temperature rise rate, and vibration spectrum drift constitute the input of the RUL prediction model; Model: A hybrid model based on random forest regression and deep residual network ResNet, outputting remaining lifespan and health index HI; Early warning mechanism: When HI<0.6, the system displays a yellow warning in the operating room; when HI<0.3, a red warning is triggered and a maintenance work order is automatically generated.

[0058] 4. Human-Computer Interaction Layer 4.1 Local Display The touchscreen uses the Qt framework to implement real-time waveform drawing, fault code pop-ups, and operation logs; Interaction logic: Clicking on the fault code will expand the fault details, including screenshots of abnormal waveforms, location coordinates, and suggested repair steps; long-pressing will directly initiate a remote diagnostic request.

[0059] 4.2 Mobile App Platform: Dual platforms including iOS and Android, using Flutter for cross-platform development; Function: Real-time monitoring dashboard: current, temperature, vibration; Historical reports: PDF export; Fault alarm push notifications: Supports offline caching; Remote authorization OTA: Two-factor authentication; Repair work order management: status flow and image upload.

[0060] 4.3 Voice / Text Interaction It integrates speech recognition and synthesis, enabling the system to return text or voice feedback when given natural language commands such as "Please check the motor temperature".

[0061] Example 2 Methods for locating electrical faults in electric staplers, such as Figure 2-3 As shown, it includes the following steps: S1. Data Acquisition and Preprocessing S1.1 High-fidelity data acquisition To achieve accurate fault diagnosis and location, the data acquisition system must meet the requirements of high dynamics and multiple physics fields.

[0062] An isolated differential probe and a high-precision Hall current sensor are used to simultaneously acquire three-phase voltage and current. The power supply voltage needs to be monitored separately. The sampling rate should be no less than 10kHz, and a resolution of 16 bits or higher is recommended to ensure the capture of nanosecond-level switching transients and microsecond-level current ripple.

[0063] An ICP-type accelerometer is used, installed in the bearing housing at both the drive and non-drive ends of the motor, to collect vibration signals in the X, Y, and Z directions. The sampling rate needs to be ≥20kHz to cover the high-frequency resonant frequencies of the bearings and gears.

[0064] Temperature sensors are placed at key points such as the motor stator windings, bearing outer rings, and power device heat sinks. Rotor temperature can be obtained indirectly using wireless telemetry or impedance-based soft measurement techniques.

[0065] High-resolution photoelectric encoders or rotary transformers are used to accurately obtain rotor position and speed, which is the basis for subsequent phase analysis.

[0066] All sensor data acquisition must be strictly synchronized by a master clock with a synchronization accuracy better than 1 microsecond to ensure strict correspondence between electrical, mechanical, and thermal signals on the time axis, laying the foundation for subsequent multi-source information fusion positioning.

[0067] S1.2 Data Preprocessing and Augmentation The raw signal contains a lot of noise, and direct analysis will seriously affect the accuracy of feature extraction.

[0068] To address the transient and non-stationary characteristics of motor signals, the Db4 or Symlets wavelet series is selected, as they exhibit good localization characteristics in the time and frequency domains.

[0069] An improved Stein unbiased risk estimation threshold or a Minimax threshold is used to adaptively soft-threshold each wavelet decomposition coefficient. This effectively removes Gaussian white noise while preserving signal edges and abrupt changes, such as current spikes during short circuits.

[0070] By employing operating condition adaptive normalization, taking current as an example, the theoretical rated current value under the given operating condition is first calculated based on the motor's real-time speed and torque commands. Then, the actual current signal is divided by this theoretical value. This processed characteristic quantity exhibits better operating condition invariance, facilitating comparisons under different loads.

[0071] To expand the training samples for CNNs, methods such as adding Gaussian noise, random time stretching, amplitude scaling, and simulating sensor drift can be used to generate labeled synthetic fault data, thereby improving the model's generalization ability.

[0072] S2. Multidimensional Feature Extraction This step aims to extract a set of features that are sensitive to and discriminative about faults from the preprocessed data.

[0073] S2.1 In-depth study of temporal features The basic characteristics are as follows: mean, effective value, peak value, peak-to-peak value, waveform factor, peak factor, and impulse factor.

[0074] The transient characteristics are as follows: Rise / fall time: Accurately calculates the rise / fall time of the current or back EMF waveform during commutation. Local short circuits or open circuits in the windings will cause changes in the electromagnetic time constant, directly affecting this parameter.

[0075] Zero-crossing distortion analysis: Precisely measures the time interval and offset of zero-crossing points. For open-circuit faults, the zero-crossing point of a specific phase will be significantly offset or disappear.

[0076] Higher-order statistics: skewness measures the asymmetry of the distribution; kurtosis measures the kurtosis of the distribution and is extremely sensitive to shock-related failures such as bearing damage.

[0077] S2.2 In-depth analysis of frequency domain characteristics Windowed interpolation FFT is used to reduce the picket fence effect and spectral leakage, and the amplitude and phase of each harmonic are accurately calculated.

[0078] For vibration signals, in addition to the fundamental frequency and its harmonics, focus on the bearing's characteristic frequencies, such as the inner ring, outer ring, rolling elements, cage, and their sidebands. Bearing failures can excite vibrations at these characteristic frequencies.

[0079] Calculate the total harmonic distortion (THD) and total demand distortion (TDD). For permanent magnet synchronous motors, stator winding asymmetry or rotor demagnetization faults will produce significant changes in harmonics of specific orders. These characteristic harmonics are more directional than THD.

[0080] S2.3 Statistical characteristics of KDE based on kernel density estimation Under healthy motor conditions, phase current and voltage data are collected over a long period during steady-state operation. A Gaussian kernel function is used to perform nonparametric estimation of the probability density function of the data, establishing a healthy fingerprint model.

[0081] During online monitoring, real-time data segments are substituted into the established healthy KDE model to calculate the log-likelihood value of the segment. A lower likelihood value indicates a greater discrepancy between the current data distribution and the healthy model, and a higher probability of anomaly. This method is particularly effective for slowly occurring performance degradation.

[0082] S3. Fault Mode Matching Based on Deep Learning and Uncertainty Measurement S3.1 Refinement of CNN Classification Network Structure Input layer: Constructs 2D feature maps from multi-sensor data as input. For example, a row can represent a point in time, and different columns can represent three-phase current, voltage, vibration channels, etc., thus visualizing the time series data.

[0083] Network structure: Convolutional layers: Design 3 to 4 convolutional layers, using small-sized kernels, to capture fault features from local to global through deep stacking. Use the ReLU activation function.

[0084] Pooling layer: Max pooling is used for downsampling to enhance the translation invariance of features.

[0085] Fully connected layer and output layer: The last layer consists of 1-2 fully connected layers. The output layer uses the Softmax function to output the probability distribution of each fault category.

[0086] Training strategy: Cross-entropy loss function combined with Adam optimizer is used. To prevent overfitting, Dropout layer is introduced into the network, and transfer learning is used to fine-tune the model pre-trained on public datasets when training data is insufficient.

[0087] S3.2 Confidence-Based Decision Making and Uncertainty Management Setting a confidence threshold of 0.85 is an engineering trade-off. Results above this threshold are considered reliable and output directly. When the highest probability (i.e., confidence level) of the CNN output falls below 0.85, it indicates that the current state may resemble multiple failure modes or represent a novel failure mode not previously learned. In this case, the system should not force an unreliable diagnosis but automatically trigger a digital twin simulation verification process. This design significantly improves the system's reliability and its ability to handle unknown situations.

[0088] S4. Simulation Verification and Root Cause Analysis Driven by Digital Twin S4.1 Construction of High-Fidelity Digital Twin Model Multiphysics coupling model: Electromagnetic Model: A 2D / 3D transient electromagnetic field model of the motor is established based on the finite element method (FEM) to accurately simulate the spatial distribution of the windings, the nonlinear magnetization characteristics of the iron core, and the magnetic field of the permanent magnet. Thermal model: Establish a lumped parameter or 3D thermal network model, considering heat sources such as copper loss, iron loss, and wind wear, as well as the heat dissipation capacity of the cooling system; Mechanical Model: Establish a multibody dynamics model of the rotor-bearing system to simulate mechanical effects such as imbalance, misalignment, and gear meshing; Model parameterization and calibration: Key parameters in the model, such as winding resistance, inductance, and bearing clearance, are designed to be configurable. During the initial system debugging phase, the model is finely calibrated using measured data under healthy conditions to ensure that the simulation and measured errors are within acceptable limits.

[0089] S4.2 Simulation Verification Process When a low-confidence case is received from the CNN, the system uses the current actual operating conditions—speed, load, and temperature—as well as key features extracted from the S2 stage, such as specific harmonic components and asymmetry, as boundary conditions and injects them into the digital twin model.

[0090] Candidate fault modes are sequentially injected into the simulation environment. Preferably, 10% of the stator A-phase windings are set to have inter-turn short circuits in the simulation model.

[0091] Run the simulation to obtain the response data of the simulation model after the fault is injected: three-phase current and vibration spectrum, and compare them comprehensively with the actual measurement data. Calculate key indicators, such as the amplitude of specific harmonics and the error of temperature field distribution. If the error between the simulation results and the actual data for a certain fault mode is within a preset range, the fault mode is confirmed, and the fault location is accurately located using the simulation model. For example, by observing the area with the most severe magnetic field distortion through simulation, the precise slot number of the short-circuit winding can be located.

[0092] S4.3 New Fault Detection and Expert System Intervention If the simulation results of all preset fault modes fail to match the actual data, the system will mark it as an unknown / novel fault. At this point, the following actions will be automatically triggered: Data Packaging and Upload: Package all the original data, feature extraction results, and simulation logs of this case, attach a preliminary analysis report, and upload them to the cloud-based expert knowledge base.

[0093] Trigger manual analysis: Notify domain experts to conduct analysis. After expert diagnosis and confirmation, the characteristics, patterns, and solutions of this new type of fault are entered into the knowledge base.

[0094] Online model evolution: This new fault sample is added to the CNN training set to initiate the incremental learning or online update process of the model, enabling the system to diagnose such new faults and achieve self-evolution.

[0095] S5. Fault location output of spatial and electrical coordinates After the fault mode is identified, it needs to be converted into location information that can be executed in engineering.

[0096] (1) Electrical phase positioning For faults such as winding short circuits and open circuits, analyzing the instantaneous phase difference of the three-phase current / voltage can pinpoint which phase the fault occurs in. Combining this with digital twin simulations of magnetic field asymmetry, the specific branch or even slot within the same phase can be located further.

[0097] (2) Mechanical positioning Based on vibration propagation: For bearing failures, by comparing the time difference or energy attenuation of impact signals received by multiple acceleration sensors, the Time Difference of Arrival (TDOA) algorithm can be used to triangulate the approximate area of ​​the faulty bearing on the mechanical structure.

[0098] Acoustic imaging: For abnormal noises caused by partial discharge or severe wear, an acoustic camera can be used to locate the sound source and visually display the fault point.

[0099] Thermal imaging location: The motor housing is scanned using an infrared thermal imager. Overheating faults will appear as abnormal hot spots, thus achieving non-contact location.

[0100] Fault report generation includes: fault code, location coordinates, severity level assessment, detailed maintenance operation guide, and software upgrade recommendations.

[0101] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. An electrical fault location system for an electric stapler, characterized in that, Comprise: A multi-modal sensing subsystem is arranged on the drive motor, power supply module, control circuit and key mechanical components of the electric anastomat for collecting electrical, thermal and mechanical multi-dimensional operation data; An edge computing unit is in communication connection with the multi-modal sensing subsystem, and is provided with a real-time sampling engine, a signal preprocessing module and an AI inference acceleration module for synchronous sampling, feature extraction and fault mode recognition of the collected data; A digital twin model is constructed in the cloud or locally and is synchronized with the data of the edge computing unit for fault simulation and verification when the fault recognition confidence is lower than a set threshold; A fault diagnosis engine is integrated in the edge computing unit and / or the cloud for fault classification and positioning based on multi-source feature fusion and a deep learning model; A human-computer interaction module is used to display fault information, receive user instructions and support remote maintenance operations.

2. The electrical fault location system for a powered stapler of claim 1, wherein, The multi-modal sensing subsystem includes current sensors, temperature sensors, acceleration sensors and optical position sensors arranged on the motor stator, motor bearing, power supply module, control circuit and overall machine body, and accesses the edge computing unit through an Ethernet-TSN bus with a timing synchronization error of not more than 1 µs.

3. The electrical fault location system for a powered stapler of claim 1, wherein, The edge computing unit uses a RISC-V AI accelerator as the core processor, is equipped with a real-time sampling engine to realize multi-channel synchronous sampling, a signal preprocessing module to complete FIR filtering and FFT operation, and an AI inference module to execute a CNN-LSTM double-flow network model with a single inference time of less than 5 ms.

4. The electrical fault location system for a powered stapler of claim 1, wherein, The digital twin model is an electromagnetic-thermal-mechanical multi-physical field coupling model constructed based on the finite element method, supports fault simulation library injection, and automatically triggers simulation verification when the confidence output by the fault diagnosis engine is lower than the threshold, and confirms the fault type and location by comparing the correlation coefficients of the simulation waveform and the actual waveform.

5. The electrical fault location system for a powered stapler of claim 1, wherein, The electrical fault positioning system further comprises a predictive maintenance module for predicting the remaining life based on a random forest and ResNet hybrid model and triggering an early warning when the health index HI is lower than the index threshold.

6. A method for locating an electrical fault in an electrically powered anastomosis device, by means of a system according to any one of claims 1-5, characterized in that, Comprise the following steps: S1. Synchronously collect electrical, thermal and mechanical multi-dimensional operation data through multi-modal sensors arranged on key parts of the electric anastomat, and perform noise reduction and normalization preprocessing on the collected raw data; S2. Extract time domain, frequency domain and statistical features based on kernel density estimation from the preprocessed data to form a multi-source feature vector; S3. Input the multi-source feature vector into a pre-trained deep learning model for fault classification and output its confidence; When the confidence is lower than the preset threshold, trigger the digital twin simulation verification process; S4. Inject the current operating condition and the multi-source feature vector into a high-fidelity digital twin model as boundary conditions for multi-physical field coupling simulation, and confirm the fault mode and locate the root cause by comparing the simulation data and the measured data; S5. Output fault positioning information including electrical phase, mechanical position or thermal image coordinates according to the confirmed fault mode, and generate a maintenance guidance report.

7. The method of claim 5, wherein, In step S1, specifically comprising: Collect three-phase electrical parameters of the drive motor using a current sensor; The mechanical vibration signal is collected by a vibration sensor; The thermal state of the key part is monitored by a temperature sensor; The displacement information of the actuator is obtained by an optical position sensor; The data collection of all sensors is synchronized by a master clock.

8. The method of claim 5, wherein, In step S2, specifically comprising: The effective value, peak value, peak-peak value, waveform factor, pulse factor, rising / falling edge time and zero-crossing distortion of the current and vibration signal are calculated; The fundamental wave, amplitude and phase of each harmonic and the total harmonic distortion rate are extracted by windowed interpolation FFT analysis of the current signal; The bearing characteristic frequency and its sideband components are extracted by spectrum analysis of the vibration signal.

9. The method of claim 5, wherein, After step S5, a predictive maintenance step is further included: Based on the cumulative running time, load cycle, temperature rise rate and vibration spectrum drift characteristics, the residual service life and health index of the equipment are predicted by a hybrid model combining random forest and residual network; when the health index is lower than the index threshold, the system issues a warning.

10. The method of claim 5, wherein, In step S4, the high-fidelity digital twin model is an electromagnetic-thermal-mechanical multi-physical field coupling model based on the finite element method; The comparison between the simulation data and the measured data is realized by calculating the cross-correlation coefficient of the simulation waveform and the field sampling waveform; when the correlation coefficient is greater than 0.92, it is confirmed that the fault mode matches.

Images