Robot intelligent inspection system and method for permanent magnet shielding pump

Abstract

This application discloses a robotic intelligent inspection system and method for permanent magnet shielded pumps, relating to the field of intelligent inspection of industrial equipment. The method includes: an on-site inspection robot that performs anomaly identification based on multi-source state data to obtain anomaly identification results under abnormal conditions; a cloud-based intelligent management and control platform including an analysis engine module, an instruction generation and distribution module, and a human-machine interface. The analysis engine module is used to determine whether there is an emergency situation with the permanent magnet shielded pump based on pre-stored historical data and anomaly identification results. The instruction generation and distribution module is used to issue control instructions to the emergency operation execution module of the on-site inspection robot through an edge computing gateway when it is determined that there is an emergency situation with the permanent magnet shielded pump, so as to execute a preset physical intervention operation; and to generate an operation card when it is determined that there is no emergency situation with the permanent magnet shielded pump, so as to maintain the operation of the permanent magnet shielded pump. This application supports two parallel anomaly handling paths, improving the timeliness of inspection and handling.

Description

Technical Field

[0001] This application relates to the field of intelligent inspection of industrial equipment, and in particular to a robotic intelligent inspection system and method for a permanent magnet shielded pump. Background Technology

[0002] With the continuous improvement of industrial automation, permanent magnet shielded pumps have been widely used in key fields such as petroleum, chemical, pharmaceutical, and environmental protection due to their advantages of high efficiency, low noise, and leak-free operation. The operating status of this type of pump set is directly related to the safety and continuity of the entire production process. Therefore, regular and efficient inspection and maintenance of pump sets is an important part of ensuring safe industrial production.

[0003] Traditional pump unit inspection methods mainly rely on regular manual patrols, which suffers from significant issues such as high subjectivity, high labor intensity, and difficulty in data traceability. With the development of IoT technology, some locations have begun to introduce inspection robots to replace manual data collection. However, in current practical applications, even with the introduction of inspection robots, their working mode remains at the "assisting human" stage. After collecting video, infrared thermal images, and vibration data on-site according to a preset route, the inspection robot transmits the data back to the back-end monitoring center. When the system detects an anomaly through threshold judgment, it still needs to notify the on-duty management personnel. After receiving the alarm, the management personnel often need to personally rush to the site for secondary confirmation, judge the severity of the problem based on their personal experience, and then report upwards level by level, awaiting instructions from superiors.

[0004] While the above model reduces the frequency of manual inspections to some extent, it does not fundamentally solve the problem of timeliness in emergency response. Furthermore, due to the significant information gap between front-end data collection and back-end decision-making, there is often a long time window between the occurrence of an anomaly and the issuance of the final instruction, which can easily lead to small hidden dangers evolving into major malfunctions. Summary of the Invention

[0005] The purpose of this application is to provide a robotic intelligent inspection system and method for permanent magnet shielded pumps, which can solve the problems mentioned above, and can build a complete closed loop from anomaly identification to emergency judgment, and then to physical intervention or work order dispatch, thereby improving the timeliness and intelligence level of inspection and handling.

[0006] To achieve the above objectives, this application provides the following solution: In the first aspect, this application provides a robotic intelligent inspection system for a permanent magnet shielded pump, including an on-site inspection robot, an edge computing gateway, and a cloud-based intelligent management and control platform; The on-site inspection robot is used to collect multi-source status data during the operation of the permanent magnet shielded pump, and to perform anomaly identification based on the multi-source status data to obtain anomaly identification results under abnormal conditions. The cloud-based intelligent management and control platform includes an analysis engine module, an instruction generation and distribution module, and a human-machine interface. The analysis engine module receives the anomaly identification results through an edge computing gateway and determines whether the permanent magnet shielded pump is in an emergency based on pre-stored historical data and the anomaly identification results. The instruction generation and distribution module, when determining that the permanent magnet shielded pump is in an emergency, issues control instructions to the emergency operation execution module of the on-site inspection robot through the edge computing gateway to execute preset physical intervention operations. When determining that the permanent magnet shielded pump is not in an emergency, it predicts multi-source state data within a preset future time range based on time series analysis and generates operation cards based on the prediction results. The operation cards include operations to be performed after the prediction results exceed a prediction threshold. The human-computer interaction interface is used to display the operation card.

[0007] In one embodiment, the on-site inspection robot includes a robot body, a multi-source data acquisition module, an edge computing module, and an emergency operation execution module. The multi-source data acquisition module is mounted on the robot body and is used to acquire the multi-source state data; the edge computing module is connected to the multi-source data acquisition module and is used to perform anomaly identification based on the multi-source state data to obtain anomaly identification results under abnormal conditions; the emergency operation execution module is used to execute preset physical intervention operations under the control of control commands issued by the command generation and issuance module.

[0008] In one embodiment, the multi-source state data includes visual data, temperature data, abnormal noise data, and vibration data. The visual data is acquired by a high-definition visible light camera, the temperature data is acquired by an infrared thermal imager, the abnormal noise data is acquired by a microphone, and the vibration data is acquired by a vibration sensor.

[0009] In one embodiment, the emergency operation execution module is a robotic arm, which is used to press the emergency stop button when it receives a control command from the command generation and distribution module. Alternatively, the emergency operation execution module may be an infrared controller, which is used to remotely control the intelligent valve of the permanent magnet shielded pump in the pump room to close when it receives a control command from the command generation and distribution module.

[0010] In one embodiment, the edge computing module is further configured to generate a preliminary warning command to the alarm of the robot body after obtaining the anomaly identification result under abnormal conditions, thereby realizing a preliminary light warning.

[0011] In one embodiment, the edge computing module adopts a heterogeneous computing architecture consisting of a CPU and an NPU, wherein the CPU is used to execute general computing tasks, and the NPU is used to perform anomaly identification based on the multi-source state data to obtain anomaly identification results under abnormal conditions.

[0012] In one embodiment, the on-site inspection robot further includes a QoS-level transmission module, which is used to select the corresponding transmission channel and priority for data transmission based on the urgency of the anomaly identification results.

[0013] In one embodiment, the analysis engine module includes a fault diagnosis model and an energy consumption analysis model; The fault diagnosis model is used to perform fault diagnosis based on the received anomaly identification results and pre-stored historical data, and outputs the fault probability and remaining life prediction to determine whether the permanent magnet shielded pump is in an emergency. The energy consumption analysis model is used to analyze the operating energy efficiency of the permanent magnet shielded pump and generate optimization suggestions.

[0014] In one embodiment, the cloud-based intelligent management and control platform further includes a data middle platform, which is used to store the inspection data of the on-site inspection robot and the records of manual handling.

[0015] Secondly, this application also provides a robotic intelligent inspection method for a permanent magnet shielded pump, comprising the following steps: The on-site inspection robot collects multi-source status data of the permanent magnet shielded pump during operation, and performs anomaly identification based on the multi-source status data to obtain the anomaly identification results under abnormal conditions. Based on the pre-stored historical data and the anomaly identification results, it is determined whether there is an emergency situation with the permanent magnet shielded pump. When it is determined that there is an emergency situation with the permanent magnet shielded pump, a control command is sent to the emergency operation execution module of the on-site inspection robot to execute a preset physical intervention operation. When it is determined that there is no emergency situation with the permanent magnet shielded pump, a multi-source state data within a preset time range is predicted based on the time series analysis method, and an operation card is generated based on the prediction results. The operation card is displayed through a human-computer interaction interface.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a robotic intelligent inspection system for permanent magnet shielded pumps. By constructing a closed-loop architecture that integrates an on-site inspection robot, an edge computing gateway, and a cloud-based intelligent management platform, it can identify anomalies based on multi-source status data, obtaining anomaly identification results under abnormal conditions. Furthermore, the cloud-based intelligent management platform can directly determine emergency situations based on the anomaly identification results and pre-stored historical data. Simultaneously, by integrating an emergency operation execution module into the on-site inspection robot, the system supports two parallel handling paths when an anomaly occurs: one path directly triggers the robot to perform physical intervention operations such as remote emergency stop; the other path automatically generates operation cards and distributes them to maintenance personnel's mobile terminals for precise maintenance. The two paths can be triggered simultaneously without interference. This ensures timely remote intervention in emergency situations affecting the pump, preventing the escalation of faults due to delays in personnel arrival, and solves the technical problem of delayed emergency response in existing permanent magnet shielded pump inspection solutions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic block diagram of a robotic intelligent inspection system for a permanent magnet shielded pump according to an embodiment of this application; Figure 2 This is an overall block diagram of a robotic intelligent inspection system for a permanent magnet shielded pump according to an embodiment of this application; Figure 3 This is a schematic block diagram of a field inspection robot for a robotic intelligent inspection system of a permanent magnet shielded pump according to an embodiment of this application. Figure 4 This is a flowchart of a robotic intelligent inspection method for a permanent magnet shielded pump according to an embodiment of this application; Figure 5 This is a flowchart illustrating the overall process of a robotic intelligent inspection method for a permanent magnet shielded pump according to an embodiment of this application. Figure 6 This is a timing diagram of the automatic handling of abnormal events in the robotic intelligent inspection method for a permanent magnet shielded pump according to an embodiment of this application. Figure 7 This is a schematic diagram illustrating the principle of data management and decision support for a robotic intelligent inspection method for a permanent magnet shielded pump according to an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] The existing technical solution follows these steps when performing inspection tasks on permanent magnet shielded pumps: Inspection Task Initiation: Back-end management personnel set a scheduled task or manually issue instructions through a remote management platform to start the inspection robot. On-site Data Acquisition: The robot moves along a preset track or path, stops at the designated workstation, and uses a high-definition camera to photograph the pump's appearance and instrument readings. An infrared thermal imager collects the temperature of key components such as bearings and motors, and a microphone collects operating noise. Data Transmission and Display: The robot transmits the collected video stream and data back to the local monitoring station and remote management platform in real time via a wireless network. The on-site footage is simultaneously displayed on the monitoring center's screen. Abnormal Alarm Trigger: The software built into the local monitoring station compares the collected temperature and vibration data with preset thresholds. When the temperature at a measuring point exceeds the set value (e.g., 85℃), the system pops up an alarm window on the interface and issues an audible and visual alert. Manual On-site Verification: After seeing the alarm alert, the on-duty personnel in the control room notify nearby inspection personnel via telephone or walkie-talkie. On-site handling and reporting: Inspection personnel arrive at the site with tools, first verifying the authenticity of the alarm (ruling out sensor false alarms), then observing the specific condition of the pump (e.g., whether there are leaks, unusual odors). After confirming the problem exists, the inspection personnel report the situation to the shift leader or equipment supervisor by phone. Superior decision-making and instruction issuance: Based on the reported information and personal experience, the equipment supervisor determines whether an emergency shutdown, process parameter adjustment, or maintenance arrangement is necessary. The decision-making instruction is issued to on-site personnel via telephone or workflow software. Result recording: After handling the issue, the inspection personnel fill out a paper or electronic form, recording the fault phenomenon and handling result, and archive it for future reference.

[0022] The existing solutions described above only use inspection robots as data collection terminals, while decision analysis and disposal execution still mainly rely on manual labor. This model results in a "time cliff" between the triggering of abnormal alarms and the issuance of disposal instructions, leading to a serious lag in emergency response.

[0023] Therefore, see Figure 1 This application provides a robotic intelligent inspection system for permanent magnet shielded pumps, including an on-site inspection robot, an edge computing gateway, and a cloud-based intelligent management and control platform.

[0024] The on-site inspection robot is used to collect multi-source status data during the operation of the permanent magnet shielded pump, and to identify anomalies based on this data, obtaining anomaly identification results under abnormal conditions. In this application, the on-site inspection robot is deployed at the pump station site and serves as the system's sensing and execution terminal.

[0025] In the embodiments of this application, see Figure 3 The on-site inspection robot includes a robot body, a multi-source data acquisition module, an edge computing module, and an emergency operation execution module. The multi-source data acquisition module is set on the robot body and is used to collect multi-source status data. The edge computing module is connected to the multi-source data acquisition module and is used to perform anomaly identification based on the multi-source status data to obtain the anomaly identification results under abnormal conditions. The emergency operation execution module is used to execute preset physical intervention operations under the control of the control commands issued by the command generation and issuance module.

[0026] Specifically, the multi-source status data includes visual data, temperature data, abnormal noise data, and vibration data. Visual data is collected by a high-definition visible light camera, temperature data is collected by an infrared thermal imager, abnormal noise data is collected by a microphone, and vibration data is collected by a vibration sensor.

[0027] For example, a high-definition visible light camera is used to read instrument pointers and observe the appearance of the permanent magnet shielded pump body; an infrared thermal imager is used to monitor the temperature of bearings and motors; a microphone is used to collect abnormal operating noises of the permanent magnet shielded pump; and a vibration sensor is used to collect the vibration spectrum, vibration velocity, and vibration acceleration of the permanent magnet shielded pump body.

[0028] When a permanent magnet shielded pump malfunctions, an emergency is detected. The emergency operation execution module receives control commands to execute preset physical intervention operations. If the emergency is not addressed, cavitation will rapidly corrode the impeller, damage the bearings, and may even cause the permanent magnet shielded pump's bearings to break. In this embodiment, the emergency operation execution module is integrated into the on-site inspection robot, serving as the system's physical execution terminal. Physical intervention operations refer to direct physical control actions performed on the permanent magnet shielded pump or its associated equipment through the emergency operation execution module, including pressing the emergency stop button to cut off the control circuit and remotely closing smart valves, to achieve emergency response without the need for on-site personnel.

[0029] For example, the emergency operation execution module is a robotic arm, which is used to press the emergency stop button when it receives a control command from the command generation and distribution module; or, the emergency operation execution module is an infrared controller, which is used to remotely control the intelligent valve of the permanent magnet shielded pump in the pump room to close when it receives a control command from the command generation and distribution module.

[0030] In this embodiment, the edge computing module adopts a heterogeneous computing architecture consisting of a CPU and an NPU. The CPU is used to execute general computing tasks, and the NPU is used to perform anomaly identification based on multi-source state data to obtain anomaly identification results under abnormal conditions.

[0031] For example, a heterogeneous computing architecture consisting of a DTB-3094-H610E main controller (CPU) and an RK3588+RK1828 coprocessor (NPU) is used. The CPU is responsible for general computing tasks such as task scheduling, navigation control, and communication management, while the NPU is specifically responsible for AI model inference, such as YOLOv5 object detection, ResNet50 thermal image analysis, and LSTM audio analysis, enabling on-site data processing without uploading all data to the cloud.

[0032] The edge computing module is also used to generate preliminary warning commands to the robot's alarm system after receiving anomaly identification results in abnormal situations, thus providing initial light warnings. The edge computing module has a built-in embedded AI chip for real-time processing of audio and video data on the robot. For example, when a sudden temperature rise or abnormal noise at a specific frequency is detected, a preliminary warning can be triggered without requiring a data transmission back.

[0033] In this embodiment, the on-site inspection robot also includes a QoS-based hierarchical transmission module, used to select the corresponding transmission channel and priority for data transmission based on the urgency of the anomaly identification results. The event-level-based QoS-based transmission mechanism: The robot is internally equipped with a QoS-based hierarchical transmission module, which selects different transmission channels and priorities according to the urgency of the abnormal event: Emergency events (such as extreme temperature or leakage): Direct transmission to the cloud via a 5G URLLC ultra-reliable low-latency channel to seize transmission resources; Ordinary events: Transmitted to the edge gateway cache via the MQTT protocol, using a best-effort transmission approach; Normal data: Only the summary or feature value is uploaded, while the original data is stored locally in a rotating manner.

[0034] In this embodiment, the on-site inspection robot also includes an autonomous walking module, which includes a lidar (LIDAR, i.e., laser detection and ranging), an inertial measurement unit (IMU), and a magnetic navigation sensor, used to achieve centimeter-level positioning and path planning for the robot in complex pump rooms.

[0035] The on-site inspection robot also includes a two-way voice intercom module, which contains a microphone and a speaker for real-time communication between on-site personnel and remote experts.

[0036] The on-site inspection robot also includes a wireless communication module, which supports multi-mode communication of 5G / WiFi / industrial Ethernet to ensure real-time data transmission.

[0037] In the embodiments of this application, see Figure 2The edge computing gateway is deployed in the on-site data center as a "data aggregation point." It connects robots and the cloud platform, and is responsible for data protocol conversion and data cleansing at the edge. It also has the function of resuming interrupted transmission when the network is interrupted.

[0038] In this embodiment, the cloud-based intelligent management and control platform is deployed in the data center and serves as the "decision-making brain" of the system. The cloud-based intelligent management and control platform includes an analysis engine module, an instruction generation and distribution module, and a human-machine interface. The analysis engine module receives anomaly identification results through an edge computing gateway and determines whether the permanent magnet shielded pump is in an emergency based on pre-stored historical data and the anomaly identification results. The instruction generation and distribution module, when determining that the permanent magnet shielded pump is in an emergency, issues control instructions to the emergency operation execution module of the on-site inspection robot through the edge computing gateway to execute preset physical intervention operations. When determining that the permanent magnet shielded pump is not in an emergency, it predicts multi-source state data within a preset future time range based on time series analysis methods and generates operation cards based on the prediction results. The operation cards include the operations to be performed after the prediction results exceed a prediction threshold and are displayed by the human-machine interface.

[0039] Specifically, the analysis engine module includes a fault diagnosis model and an energy consumption analysis model. The fault diagnosis model is used to diagnose faults based on the received anomaly identification results and pre-stored historical data, and outputs the fault probability and remaining life prediction to determine whether there is an emergency situation for the permanent magnet shielded pump. The energy consumption analysis model is used to analyze the operating energy efficiency of the permanent magnet shielded pump and generate optimization suggestions.

[0040] The cloud-based intelligent management and control platform also includes a data platform, which stores inspection data from on-site inspection robots and manual handling records. The data platform stores inspection data uploaded by on-site inspection robots and manual handling records returned by maintenance personnel via mobile terminals. The inspection data includes various types such as structured data, images, and videos. This application links and archives the robot's automatic inspection data with manual handling results, forming a closed-loop database covering the entire lifecycle of the equipment. This database provides historical data support for the analysis engine module, enabling the system to perform fault diagnosis, remaining life prediction, and energy efficiency optimization analysis based on historical trends.

[0041] The human-machine interface in this application is deployed on a cloud-based intelligent management platform (HMI), and is provided to equipment supervisors, plant managers, and other senior management personnel in the form of a web interface or an app. This interface supports real-time 3D visualization of the permanent magnet shielded pump using digital twin technology, enabling managers to intuitively view equipment operating parameters, fault locations, and energy efficiency status.

[0042] The operation cards in this application are standardized maintenance cards automatically generated by the cloud-based intelligent management and control platform when a non-emergency situation is determined. Specifically, based on time series analysis methods (such as exponential smoothing and autoregressive prediction models), multi-source state data within a preset future time range (such as the next 10 minutes or the next 6 hours) are predicted, and the prediction results are obtained. The prediction results are compared with pre-set prediction thresholds to obtain the corresponding execution operations, and the operation cards are integrated.

[0043] In an exemplary embodiment, an exponential smoothing method is used to predict multi-source state data. Specifically, different weights are assigned to historical multi-source state data, with more recent multi-source state data having a larger weight and earlier multi-source state data having a smaller weight. Through smoothing and weighting, multi-source state data within a preset future time range can be obtained.

[0044] In one exemplary embodiment, an autoregressive prediction model is used to predict multi-source state data. Specifically, multi-source state data from the past 100 time points are collected, and differencing operations are performed to transform the non-stationary sequence into a stationary sequence. Then, an autoregressive prediction model is established to predict multi-source state data within a preset future time range.

[0045] In one exemplary embodiment, the future preset time range includes two modes: short-term (10 minutes) and long-term (6 hours). When the future preset time range is 10 minutes, the prediction result after 10 minutes is determined, and the prediction result is compared with a pre-set prediction threshold. If the prediction result is greater than the prediction threshold, maintenance personnel should take measures within 5 minutes according to the operation card. When the future preset time range is 6 hours, the prediction result after 6 hours is determined, and the prediction result is compared with a pre-set prediction threshold. If the prediction result is greater than the prediction threshold, the content on the operation card is added to the daily maintenance plan, and maintenance personnel should take response measures within 4 hours according to the operation card.

[0046] In one exemplary embodiment, vibration velocity (vibration data), vibration acceleration (vibration data), vibration spectrum (vibration data), and bearing temperature (temperature data) within a preset future time range are determined, and corresponding prediction thresholds are set. In this embodiment, the prediction thresholds include a vibration velocity threshold (range [4.5, 7.0) mm / s), a vibration acceleration threshold (range [0.5, 1.0) gE), a vibration spectrum threshold (second harmonic component in the spectrum > 200% of the reference value), and a bearing temperature threshold ([80, 90) °C). When the vibration velocity within the preset future time range exceeds the corresponding vibration velocity threshold, the anchor bolt torque is checked; when the vibration acceleration within the preset future time range exceeds the corresponding vibration acceleration threshold, lubricating oil is added; when the vibration spectrum within the preset future time range exceeds the corresponding vibration spectrum threshold, the coupling is re-inspected; when the bearing temperature within the preset future time range exceeds the corresponding bearing temperature threshold, the heat sink is cleaned and the cooling system is checked. If the prediction result does not exceed the corresponding prediction threshold, no operation is performed.

[0047] In one exemplary embodiment, the operation card is displayed via a human-machine interface, and maintenance personnel perform specific operations based on the interface. Furthermore, depending on the actual situation, the cloud-based intelligent management platform can also connect to the maintenance personnel's AR glasses and transmit the operation card to their AR smart glasses. The AR smart glasses, guided by coordinates, use virtual arrows to precisely indicate the operation location (e.g., which specific bearing) within the maintenance personnel's field of vision and display the operation to be performed (e.g., adding 15g of lubricating oil). After the maintenance personnel complete the operation, the AR glasses confirm via voice or gesture and automatically transmit the result. If AR terminals are not supported, the cloud-based intelligent management platform can also send the operation card to the maintenance personnel's mobile app, which includes checkboxes, parameter input boxes, and a photo upload button.

[0048] After the maintenance personnel complete the operation according to the requirements of the operation card, they will send the executed operation back to the cloud-based intelligent management and control platform and archive it in the history record for subsequent processing effect evaluation and dynamic optimization of prediction threshold setting.

[0049] In one example, the multi-source data fusion analysis method executed on the on-site inspection robot includes: temperature analysis: ResNet50-based heatmap feature extraction + cosine similarity comparison; vibration analysis: FFT spectrum transformation + SVM classifier for bearing condition diagnosis; audio analysis: MFCC feature extraction + LSTM time-series network for abnormal sound detection; visual analysis: YOLOv5 object detection + PP-OCRv3 text recognition for instrument readings. The cloud-based XGBoost deep diagnostic and remaining life prediction method: associates historical trend data from the InfluxDB time-series database with maintenance and replacement records from the MySQL relational database; calls the XGBoost gradient boosting tree model to calculate the failure probability and remaining life based on the multi-source data; outputs structured diagnostic results: failure type, confidence level, remaining life, and recommended measures.

[0050] In summary, the inspection robot in this application not only includes multi-source data acquisition capabilities but also integrates an emergency operation execution module, constructing a two-way command closed loop from the robot → edge computing gateway → cloud-based intelligent management and control platform → robot / mobile terminal, rather than a one-way data flow. When an anomaly occurs, the system supports two parallel handling paths: autonomous robot execution (remote emergency stop) and precise manual maintenance (work order dispatch), both of which can be triggered simultaneously without interference.

[0051] This application integrates an emergency operation execution module within the inspection robot, establishing a command channel from the cloud platform to the robot, and employing 5G URLLC ultra-reliable low-latency communication to transmit emergency commands. Because the robot possesses remote execution capabilities, and cloud commands can bypass manual intervention and directly reach the robot, in the event of emergencies such as excessive temperature or leaks, the administrator can click "Emergency Stop" on the HMI interface. The command will reach the robot within 10 seconds and activate the relay, cutting off the pump's control circuit. This reduces emergency response time from hours to seconds, nipping accidents in the bud and preventing major equipment damage. Thus, it solves the response delay problem caused by existing technologies relying on manual intervention, significantly improving the inherent safety level of industrial sites.

[0052] Furthermore, this application deploys an analysis engine module on a cloud platform, runs an XGBoost gradient boosting tree model, and combines historical trends from a time-series database (InfluxDB) with maintenance records from a relational database (MySQL) to output accurate structured diagnostic results. The cloud-based intelligent management platform can acquire first-hand data such as high-definition images, vibration waveforms, infrared thermal images, and audio spectra uploaded by remote inspection robots, and can retrieve historical data and maintenance records from the past 30 days for comparison. Therefore, what senior managers see on the HMI interface is not a vague "vibration anomaly," but rather "fault prediction." For example, if the radial clearance between the sliding bearing and the bushing exceeds 0.25mm, the recommendation is to replace the bearing and the corresponding bushing, with an accurate prediction of a remaining lifespan of 72 hours. Thus, deep diagnosis based on multi-source data fusion is achieved, providing senior managers with quantified fault probabilities and remaining lifespan predictions, avoiding attenuation and distortion during information transmission, and improving the accuracy of decision-making.

[0053] Furthermore, this application constructs a data platform that automatically links and archives the robot's inspection data with the handling results transmitted from the mobile terminal, forming a complete equipment lifecycle database. Because the handling results transmitted from the mobile terminal (such as "bearing replaced, old parts severely worn") and the abnormal events reported by the robot are associated with the same device ID and timestamp, and the data platform adopts an InfluxDB+MySQL hybrid storage architecture, which can store both time-series data and structured records, when similar vibration characteristics are detected again, the system can automatically retrieve the previous maintenance records, analyze the maintenance effect, and optimize the diagnostic model.

[0054] Based on the same inventive concept, this application also provides a robotic intelligent inspection method for a permanent magnet shielded pump. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the robotic intelligent inspection method for permanent magnet shielded pumps provided below can be found in the limitations of the robotic intelligent inspection system for permanent magnet shielded pumps described above, and will not be repeated here.

[0055] See Figure 4 This application provides a robotic intelligent inspection method for a permanent magnet shielded pump, comprising the following steps: S100: Collects multi-source status data of the permanent magnet shielded pump during operation through on-site inspection robot, and performs anomaly identification based on multi-source status data to obtain anomaly identification results under abnormal conditions; S200: Based on pre-stored historical data and anomaly identification results, determine whether there is an emergency situation with the permanent magnet shielded pump. When it is determined that there is an emergency situation with the permanent magnet shielded pump, issue control commands to the emergency operation execution module of the on-site inspection robot to execute the preset physical intervention operation. When it is determined that there is no emergency situation with the permanent magnet shielded pump, predict the multi-source state data within a preset time range based on the time series analysis method, and generate an operation card based on the prediction results. S300: The operation card is displayed through the human-computer interaction interface.

[0056] See Figure 5 , Figure 6 and Figure 7 In a specific example, the steps are as follows: Autonomous Inspection and Data Acquisition: The on-site inspection robot scans the target permanent magnet shielded pump point by point along a planned path according to a preset timed task or manual triggering from the background. The surface temperature, vibration waveform, operating noise, and instrument readings of the pump body are collected synchronously through a multi-source data acquisition module. The collected data first flows into the edge computing module.

[0057] Specifically, the multi-source data acquisition module adopts a synchronous triggering mechanism. After the on-site inspection robot arrives at the preset inspection area (e.g., 0.5m away from the bearing end cover of the permanent magnet shielded pump), it triggers a transistor-transistor logic (TTL) level synchronization pulse. The TTL level synchronization pulse simultaneously activates the high-definition visible light camera, infrared thermal imager, microphone, and vibration sensor to acquire multi-source status data. The synchronous triggering mechanism ensures that visual data, temperature data, abnormal noise data, and vibration data are aligned under the same timestamp and the same spatial reference system.

[0058] When acquiring visual and temperature data, the high-definition visible light camera and infrared thermal imager perform optical axis conjugate calibration, so that the pixels of the visual data and temperature data can correspond point by point, which facilitates subsequent analysis.

[0059] When collecting abnormal noise data, a micro-electro-mechanical systems (MEMS) digital microphone array (e.g., four microphones arranged in a rectangle) is used, and a beamforming algorithm is employed to enhance the sound source signal from the direction of the permanent magnet shielded pump, thereby suppressing the environmental noise of the permanent magnet shielded pump (such as noise from other equipment and ventilation fans).

[0060] When collecting vibration data, an integrated electronic piezoelectric (IEPE) accelerometer is used. It is fixed to the end of the flexible probe of the on-site inspection robot by magnetic attraction or thread. The flexible probe extends and adheres to the measuring point of the pump body with a constant pressure (e.g., 2N) to collect vibration acceleration signals in three orthogonal directions (X: radial horizontal, Y: radial vertical, Z: axial). The sampling rate is set to 25.6kHz to meet the detection requirements of bearing fault characteristic frequencies (up to 10kHz).

[0061] Real-time edge-side analysis and preprocessing: The lightweight AI model built into the edge computing module performs real-time data analysis. The working principle is as follows: Vibration waveforms and abnormal noise data are converted into a spectrum using a Fast Fourier Transform, which is then compared with the built-in "normal operating spectrum." If the data is normal, only the summary data is packaged and uploaded to the cloud platform for long-term storage.

[0062] Specifically, the processing procedure of the edge computing module is as follows: First, the abnormal noise data is conditioned and framed. Specifically, the noise (audio spectrum) first passes through an anti-aliasing filter (cutoff frequency 0.4 times the sampling rate), and then undergoes analog-to-digital conversion. Second, since the noise from the permanent magnet shielded pump is a non-stationary signal, it needs to be divided into short-time stationary frames with a frame length of 4096 sampling points (approximately 85ms) and a frame shift of 2048 sampling points (50% overlap) to reduce energy abrupt changes between frames. Third, to eliminate spectral leakage caused by framing, each frame of data... Add a window (multiply by a Hanning window) The formula for calculating the Hanning window is as follows: ; in, This is the sampling point number.

[0063] The signal after windowing is .

[0064] A Fast Fourier Transform (FFT) is performed on the windowed signal using the radix-2 Cooley-Tukey FFT algorithm. This transforms the 4096 points of the windowed signal, mapping it from the time domain to the frequency domain. The calculation formula is as follows: ; in, Output as a complex number, representing the first... The amplitude and phase at each frequency point.

[0065] The power spectrum of a single frame is obtained by taking the square of the modulus of the signal mapped to the frequency domain. , To obtain smooth and stable spectral characteristics, an exponentially weighted moving average was performed on the power spectrum of eight consecutive frames (approximately 0.34 seconds) to obtain the final spectrum data used for analysis. Based on the physical characteristics of the permanent magnet shielded pump, the total energy of three key frequency bands in the spectrum data was extracted as fault characteristics. These three key frequency bands include a low-frequency band, a mid-frequency band, and a high-frequency band. The low-frequency band (0-1kHz) mainly reflects the pump speed and impeller imbalance; the mid-frequency band (1kHz-8kHz) mainly reflects the fault characteristic frequencies of the bearing inner and outer rings; and the high-frequency band (8kHz-20kHz) mainly reflects the lubrication state and ultrasonic signals in the early stages of cavitation.

[0066] The vibration data was then processed. First, signal processing and analog-to-digital conversion were performed on the vibration data. The charge signal output by the vibration sensor was excited by a constant current source, amplified, and filtered with anti-aliasing (cutoff frequency 12.8kHz), and then converted into a discrete digital signal at a sampling rate of 25.6kHz. This sampling rate satisfies the Nyquist sampling theorem and can cover the analysis requirements of bearing fault characteristic frequencies (up to approximately 10 kHz).

[0067] Unlike abnormal noise data, vibration data exhibits quasi-stationary characteristics when the pump speed is stable, eliminating the need for frame-by-frame processing. An integer-cycle sampling strategy is employed: 4096 sampling points (approximately 0.16 seconds) are continuously collected as one analysis sample, with a sample length covering at least 10 revolutions per second (approximately 0.0207 seconds based on a pump speed of 2900 rpm). The sampled data (discrete digital signal) is subtracted by the mean to eliminate the DC component; the calculation formula is as follows: ; in, To eliminate the DC component from the discrete digital signal, The number of sampling points. , Number the sampling points. This is a loop index variable used to iterate through the sampling points during the summation process. .

[0068] Eliminating the DC component removes the influence of vibration sensor zero drift and gravitational acceleration components, allowing subsequent spectral analysis to focus on the alternating vibration components. A radix-2 FFT algorithm is performed on the discrete digital signal after DC component elimination to convert the time-domain vibration signal (discrete digital signal) into a frequency-domain spectrum. The calculation formula is as follows: ; in, The frequency domain spectrum (frequency domain representation) is the result of performing a Fourier transform on the time domain vibration signal. For is the imaginary unit, This indicates that the FFT output is a complex number. For frequency domain frequency index, .

[0069] The effective value of vibration velocity is extracted from the frequency domain spectrum. The formula for calculating the effective value of vibration velocity is as follows: ; The effective value of vibration velocity characterizes vibration energy.

[0070] Calculate the rotational frequency based on the rated speed of the permanent magnet shielded pump. If the spectral amplitude at frequency f = the turning frequency in the spectrum This reflects impeller imbalance and shaft bending; if the frequency f =2 × spectral amplitude at the transition frequency This indicates that the coupling is misaligned.

[0071] Anomaly Triggering and Instant Reporting: When the edge computing module identifies anomalies (such as a temperature slope > 5℃ / min, or the appearance of a specific friction spectrum), it immediately uploads a high-priority message containing the precise location, fault type, and original data packet to the cloud-based intelligent management and control platform via the 5G network, while simultaneously triggering the robot's local warning lights.

[0072] Cloud-based intelligent diagnosis and decision support: After receiving abnormal data, the analysis engine module of the cloud-based intelligent management and control platform retrieves the historical medical records of the permanent magnet shielded pump from the data center. Working principle: The current vibration data is compared with the vibration data before the last maintenance to calculate the fault matching degree. Human-machine interaction: After logging into the human-machine interface, senior management personnel will see not only the word "alarm," but also "fault prediction," indicating that the radial clearance between the sliding bearing and the bushing exceeds 0.25mm, suggesting replacement of the bearing and corresponding bushing, with an estimated remaining life of 72 hours. Simultaneously, managers can click to view the original vibration waveform.

[0073] Two-way command issuance and closed-loop processing: This step includes two parallel processing paths: Path A (Remote Emergency Intervention): In case of an emergency (such as pump running dry or temperature exceeding the maximum allowable limit), the supervisor can click the "Emergency Stop" button through the human-machine interface. The instruction generation and distribution module will immediately forward the control instruction to the robot through the edge computing gateway. The robot's emergency operation execution module will then press the emergency stop button or cut off the control circuit via its robotic arm. The entire process is completed within 10 seconds without the need for personnel to be present.

[0074] Path B (Precise Manual Maintenance): Unless it is an emergency, the instruction generation and issuance module uses time series analysis to predict multi-source status data within a preset time range in the future, and generates operation cards based on the prediction results. The operation is executed when the prediction result of the operation card exceeds the prediction threshold. The executed operation includes information such as "maintenance location, maintenance phenomenon, maintenance handling measures, and tools to be carried", and is accurately pushed to the nearest and idle mobile terminal of the maintenance personnel.

[0075] On-site handling and data feedback: After receiving the operation card, maintenance personnel arrive at the site with their tools to handle the issue. Upon completion, they fill out the handling result (e.g., bearing replaced, old part severely worn) via mobile terminal and upload a photo. This information is transmitted back to the cloud platform, automatically linked and archived with the robot's inspection data, forming a complete closed-loop record.

[0076] Data accumulation and model iteration: The data platform can use the current "abnormal data + handling results" as a new sample, automatically package it and transmit it to the analysis engine module for incremental training, and continuously optimize the accuracy of the model.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A robotic intelligent inspection system for a permanent magnet shielded pump, characterized in that, This includes on-site inspection robots, edge computing gateways, and cloud-based intelligent management and control platforms; The on-site inspection robot is used to collect multi-source status data during the operation of the permanent magnet shielded pump, and to perform anomaly identification based on the multi-source status data to obtain anomaly identification results under abnormal conditions. The cloud-based intelligent management and control platform includes an analysis engine module, an instruction generation and distribution module, and a human-machine interface. The analysis engine module receives the anomaly identification results through an edge computing gateway and determines whether the permanent magnet shielded pump is in an emergency based on pre-stored historical data and the anomaly identification results. The instruction generation and distribution module, when determining that the permanent magnet shielded pump is in an emergency, issues control instructions to the emergency operation execution module of the on-site inspection robot through the edge computing gateway to execute preset physical intervention operations. When determining that the permanent magnet shielded pump is not in an emergency, it predicts multi-source state data within a preset future time range based on time series analysis and generates operation cards based on the prediction results. The operation cards include operations to be performed after the prediction results exceed a prediction threshold. The human-computer interaction interface is used to display the operation card.

2. The robotic intelligent inspection system for permanent magnet shielded pumps according to claim 1, characterized in that, The on-site inspection robot includes a robot body, a multi-source data acquisition module, an edge computing module, and an emergency operation execution module. The multi-source data acquisition module is mounted on the robot body and is used to acquire the multi-source state data; the edge computing module is connected to the multi-source data acquisition module and is used to perform anomaly identification based on the multi-source state data to obtain anomaly identification results under abnormal conditions. The emergency operation execution module is used to execute preset physical intervention operations under the control of the control commands issued by the command generation and issuance module.

3. The robotic intelligent inspection system for permanent magnet shielded pumps according to claim 2, characterized in that, The multi-source status data includes visual data, temperature data, abnormal noise data, and vibration data. The visual data is collected by a high-definition visible light camera, the temperature data is collected by an infrared thermal imager, the abnormal noise data is collected by a microphone, and the vibration data is collected by a vibration sensor.

4. The robotic intelligent inspection system for permanent magnet shielded pumps according to claim 3, characterized in that, The emergency operation execution module is a robotic arm, which is used to press the emergency stop button when it receives a control command from the command generation and sending module. Alternatively, the emergency operation execution module may be an infrared controller, which is used to remotely control the intelligent valve of the permanent magnet shielded pump in the pump room to close when it receives a control command from the command generation and distribution module.

5. The robotic intelligent inspection system for permanent magnet shielded pumps according to claim 2, characterized in that, The edge computing module is also used to generate a preliminary warning command to the alarm of the robot body after obtaining the abnormal identification result under abnormal conditions, so as to realize the preliminary light warning.

6. The robotic intelligent inspection system for permanent magnet shielded pumps according to claim 2, characterized in that, The edge computing module adopts a heterogeneous computing architecture consisting of a CPU and an NPU. The CPU is used to execute general computing tasks, and the NPU is used to perform anomaly identification based on the multi-source state data to obtain anomaly identification results under abnormal conditions.

7. The robotic intelligent inspection system for a permanent magnet shielded pump according to claim 1, characterized in that, The on-site inspection robot also includes a QoS-based transmission module, which is used to select the corresponding transmission channel and priority for data transmission based on the urgency of the anomaly identification results.

8. The robotic intelligent inspection system for permanent magnet shielded pumps according to claim 1, characterized in that, The analysis engine module includes a fault diagnosis model and an energy consumption analysis model; The fault diagnosis model is used to perform fault diagnosis based on the received anomaly identification results and pre-stored historical data, and outputs the fault probability and remaining life prediction to determine whether the permanent magnet shielded pump is in an emergency. The energy consumption analysis model is used to analyze the operating energy efficiency of the permanent magnet shielded pump and generate optimization suggestions.

9. The robotic intelligent inspection system for a permanent magnet shielded pump according to claim 1, characterized in that, The cloud-based intelligent management and control platform also includes a data middle platform, which is used to store the inspection data of the on-site inspection robot and the records of manual handling.

10. A robotic intelligent inspection method for a permanent magnet shielded pump, characterized in that, Includes the following steps: The on-site inspection robot collects multi-source status data of the permanent magnet shielded pump during operation, and performs anomaly identification based on the multi-source status data to obtain the anomaly identification results under abnormal conditions. Based on the pre-stored historical data and the anomaly identification results, it is determined whether there is an emergency situation with the permanent magnet shielded pump. When it is determined that there is an emergency situation with the permanent magnet shielded pump, a control command is sent to the emergency operation execution module of the on-site inspection robot to execute a preset physical intervention operation. When it is determined that there is no emergency situation with the permanent magnet shielded pump, a multi-source state data within a preset time range is predicted based on the time series analysis method, and an operation card is generated based on the prediction results. The operation card is displayed through a human-computer interaction interface.

Images