Lightweight industrial equipment fault diagnosis system and method based on internet of things and ai

Abstract

This invention belongs to the field of intelligent diagnostic technology for industrial equipment, specifically relating to a lightweight industrial equipment fault diagnosis system and method based on the Internet of Things and AI. The system includes an edge diagnostic unit and an optional cloud collaboration unit. The edge diagnostic unit adopts a top-down five-layer architecture, namely a data access layer, a data storage layer, a model diagnostic layer, a report generation layer, and a protocol push layer. Each layer realizes data flow and functional collaboration through standardized interfaces to complete local full-process closed-loop diagnosis. The cloud collaboration unit and the edge diagnostic unit collaborate asynchronously through a lightweight communication protocol, undertaking incremental model training and batch equipment management tasks. This invention has a high degree of integration, strong real-time performance, and strong protocol compatibility, enabling accurate and rapid diagnosis of industrial equipment faults and a closed-loop operation and maintenance process.

Description

Technical Field

[0001] This invention belongs to the field of intelligent diagnostic technology for industrial equipment, specifically relating to a lightweight industrial equipment fault diagnosis system and method based on the Internet of Things and AI. Background Technology

[0002] The continuous and stable operation of industrial equipment directly determines industrial production efficiency, operating costs, and production safety, occupying a core position in the fields of Industrial Internet of Things (IIoT), intelligent manufacturing, and equipment operation and maintenance. Existing industrial equipment fault diagnosis technologies are mainly divided into three categories: first, expert system-based diagnostic technologies, which rely on human experience to build rule bases, resulting in poor adaptability and difficulty in updating; second, traditional IoT monitoring systems, which can only achieve data collection and simple alarms, lacking in-depth analysis capabilities; and third, artificial intelligence-based diagnostic technologies, which mostly employ complex deep learning models, requiring high hardware computing power and incurring high deployment costs, making them unsuitable for the lightweight deployment needs of small diagnostic boxes.

[0003] Existing industrial equipment fault diagnosis solutions have a core flaw: 1. Limited data acquisition dimensions, using only a single type of sensor, resulting in low accuracy in fault identification and a high rate of false positives and false negatives; 2. Low integration level, separation of data processing and diagnostic models, requiring additional independent computing terminals, poor deployment and scenario adaptability, and failure to achieve local closed loop; 3. The system has weak interoperability, lacks the ability to push multiple industrial protocols, and cannot interconnect with industrial configuration software, MES systems, etc., making it difficult to form a closed loop for the entire process of diagnosis, early warning, handling, and review. The above-mentioned defects make it impossible for the existing solution to balance diagnostic accuracy, real-time performance, cost, and scenario adaptability, and it is difficult to meet the core requirements of modern industry for lightweight, localized, highly reliable, and highly adaptable equipment diagnostic systems. Summary of the Invention

[0004] To address the challenges of low accuracy in fault identification, low integration, difficulty in achieving localized closed-loop systems, and weak system interoperability in modern industrial equipment diagnostic systems, this invention provides a lightweight industrial equipment fault diagnosis system and method based on the Internet of Things and AI. This system features high integration, strong real-time performance, and strong protocol compatibility, enabling accurate and rapid fault diagnosis and a closed-loop operation and maintenance process for industrial equipment.

[0005] The technical solution of the present invention is as follows: The lightweight industrial equipment fault diagnosis system based on IoT and AI includes an edge diagnostic unit and an optional cloud collaboration unit. The edge diagnostic unit adopts a top-down five-layer architecture, consisting of a data access layer, a data storage layer, a model diagnostic layer, a report generation layer, and a protocol push layer. Each layer achieves data flow and functional collaboration through standardized interfaces, completing a local full-process closed-loop diagnosis. The cloud collaboration unit asynchronously collaborates with the edge diagnostic unit through a lightweight communication protocol, undertaking incremental model training and batch equipment management tasks.

[0006] Furthermore, the data access layer includes a protocol adaptation module and a format conversion module; The protocol adaptation module pre-integrates MQTT, HTTP, Modbus TCP / RTU, OPC UA, and RS485 / 232 communication protocols, and automatically identifies the access protocol through data feature identifier bits; The format conversion module converts multi-source heterogeneous raw data into standardized device data and filters out invalid data. The standardized equipment data includes timestamps, equipment identifiers, data types, numerical values, and units. The invalid data includes null values ​​and garbled characters.

[0007] Furthermore, the data storage layer includes a local database module and a storage strategy management module; The local database module uses a lightweight embedded database to store standardized device data, diagnostic results, and model call logs. The storage strategy management module executes daily storage and automatic cleanup rules, automatically deleting raw data that has reached the preset retention period at midnight every day, and only retaining diagnostic logs and model records that have not exceeded the preset retention period. It also supports specifying training data for temporary storage.

[0008] Furthermore, the model diagnostic layer includes a model management module, a real-time inference module, a log recording module, and a training update module; The model management module has a built-in model storage pool, which stores models in PKL format, periodically synchronizes model configuration files from the cloud collaboration unit, and updates the model parameters of the model storage pool. The real-time inference module, based on the standardized equipment data output by the data access layer and the associated equipment identifiers and operating condition information, matches the corresponding AI diagnostic model from the model storage pool, completes fault diagnosis inference locally, and outputs the fault type, level and confidence level. The log recording module fully records the model call time, device identifier, input data characteristics, diagnostic results and inference time, forming a structured call log, which is stored in the data storage layer; The training update module supports four update modes: online training, online update, offline manual update, and model parameter visualization tuning.

[0009] Furthermore, the four update modes of the training update module are as follows: Online training mode: Receives training instructions for a specified date from the user via the terminal, extracts standardized device data for the corresponding date from the data storage layer or temporary storage area, and encrypts and sends the data back to the cloud collaborative unit via API, whereby the cloud collaborative unit completes incremental model training; Online update mode: After the cloud-based collaborative unit completes model training, it generates an optimized model parameter package and pushes it to the edge diagnostic unit. The training update module automatically verifies the integrity and compatibility of the model parameter package. After the verification is passed, it automatically replaces the parameters of the old model and completes the online model update. The whole process does not require downtime and does not affect the normal operation of diagnostic services. Offline manual update mode: Supports importing offline model parameter packages. Users can update the model through the local interactive interface of the edge diagnostic unit. The training update module completes the manual update after verifying that the offline model parameter package is correct. Model parameter visualization tuning mode: Provides a visualization parameter tuning interface, allowing users to adjust the core parameters of the model according to the real-time operating conditions of the equipment. The tuned parameters take effect in real time, and the tuning effect can be quickly verified through diagnostic results feedback.

[0010] Furthermore, the report generation layer includes a knowledge base module, a report generation module, and a format conversion module; The knowledge base module has a built-in local knowledge base, which is used to store the mapping relationship between industrial equipment fault types and maintenance measures. The report generation module automatically matches the corresponding entries in the local knowledge base based on the output of the model diagnosis layer, and generates a structured diagnosis report by combining the device identifier, diagnosis time, and operating parameters. The report is divided into a simplified version and a detailed version. The simplified version contains core fault information and emergency handling suggestions for rapid early warning. The detailed version contains equipment operation data statistics, fault source analysis, and maintenance cycle suggestions for operation and maintenance review. The format conversion module supports converting reports to JSON and PDF formats. The simplified version of the report is stored in JSON format by default to meet the needs of real-time push, while the detailed version of the report can be customized to select the storage format.

[0011] Furthermore, the protocol push layer includes a push strategy module, a protocol adaptation module, and a status feedback module; it supports selecting the push strategy and communication protocol according to the fault level, pushes diagnostic reports to the terminal, and has automatic retry and status recording functions for push failures.

[0012] Furthermore, the edge diagnostic unit is an industrial-grade diagnostic box with the following hardware configuration: an ARM Cortex-A55 quad-core processor, 4GB DDR4 memory, 8GB eMMC storage, and support for M.2 / SATA expansion; it has Gigabit Ethernet, RS485, CAN, Wi-Fi / 4G network interfaces, as well as multiple analog and digital sensor interfaces, which can be connected to vibration, temperature, electrical, process, and status industrial sensors.

[0013] Furthermore, the cloud-based collaborative unit includes a model configuration module and an incremental training module; The model configuration module manages AI diagnostic models and parameter configuration files for different industrial equipment under different operating conditions, supports model version management, and records the content and time of each model parameter update. The incremental training module is used to receive data from the edge diagnostic unit for a specified date, combine it with historical training data in the cloud, and use the incremental training algorithm to train the basic model to generate an optimized model parameter package.

[0014] A fault diagnosis method for lightweight industrial equipment based on IoT and AI includes the following steps: Step 1: The edge diagnostic unit adaptively accesses industrial equipment operation data through multiple protocols and converts it into standardized equipment data; Step 2: Store standardized equipment data in a local lightweight database, automatically clean up raw data that exceeds the preset retention time, and mark temporarily stored training data; Step 3: Synchronize the cloud-based collaborative unit or call the local model parameters, match the lightweight AI model to complete local real-time inference, output diagnostic results and record the call log; Step 4: Based on the diagnostic results, match the local knowledge base to generate a tiered diagnostic report and complete the format conversion; Step 5: Select the protocol and policy according to the fault level, push the report to the target terminal and monitor the status; Step 6: Update the model via cloud-based collaborative units online, locally offline, or in parameter tuning mode; Step 7: Iterate collaboratively between edge and cloud, or manually update the model and local knowledge base locally.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) Local closed loop, extremely simple deployment: The innovative five-layer hierarchical architecture is integrated into the lightweight diagnostic box, which does not require an external computing terminal, realizing a local closed loop from data collection to result output, reducing deployment costs and operation and maintenance complexity.

[0016] (2) Multi-source adaptation and accurate diagnosis: Supports multi-protocol and multi-dimensional sensor data access, standardized equipment data processing combined with lightweight AI model, inference latency ≤500ms, and significantly reduces fault misjudgment rate and missed judgment rate.

[0017] (3) Flexible updates and scenario adaptation: It supports four model update modes: online and offline, and visual optimization, adapting to industrial scenarios with no network and weak network, and balancing diagnostic accuracy and deployment flexibility.

[0018] (4) Full-chain collaboration and efficient operation and maintenance: Multi-protocol push to connect with existing industrial systems, automatically generate hierarchical reports and push them accurately, forming a closed loop of diagnosis, early warning, handling and review, reducing operation and maintenance response time by more than 50%.

[0019] (5) Flexible expansion and continuous optimization: Supports independent operation of a single device and centralized cloud management of multiple devices, incremental model training and dynamic updates of the knowledge base, and adapts to equipment in multiple industries and new fault scenarios. Attached Figure Description

[0020] Figure 1 This is a functional diagram of the system in an embodiment of the present invention; Figure 2 Flowchart for edge diagnostic unit processing. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0022] See Figure 1 A lightweight industrial equipment fault diagnosis system based on the Internet of Things and AI includes an edge diagnostic unit and an optional cloud collaboration unit. The edge diagnostic unit adopts a top-down five-layer architecture, namely a data access layer, a data storage layer, a model diagnostic layer, a report generation layer, and a protocol push layer. Each layer realizes data flow and functional collaboration through standardized interfaces to complete local full-process closed-loop diagnosis. The cloud collaboration unit and the edge diagnostic unit collaborate asynchronously through a lightweight communication protocol to undertake incremental model training and batch equipment management tasks.

[0023] Data access layer The data access layer includes a protocol adaptation module and a format conversion module; the protocol adaptation module pre-integrates communication protocols such as MQTT, HTTP, Modbus TCP / RTU, OPCUA, and RS485 / 232, and automatically identifies the access protocol through data feature identifier bits; The format conversion module converts multi-source heterogeneous raw data into standardized device data and filters out invalid data such as null values ​​and garbled characters. The standardized equipment data includes timestamps, equipment identifiers, data types, numerical values, and units.

[0024] Data storage layer The data storage layer includes a local database module and a storage strategy management module; the local database module uses a lightweight embedded database to store standardized device data, diagnostic results, and model call logs. The storage strategy management module executes daily storage and automatic cleanup rules, automatically deleting raw data that has reached the preset retention period at midnight every day, and only retaining diagnostic logs and model records that have not exceeded the preset retention period. It also supports specifying training data for temporary storage.

[0025] Model diagnostic layer The model diagnostic layer includes a model management module, a real-time inference module, a log recording module, and a training update module; The model management module has a built-in model storage pool, which stores models in PKL format, periodically synchronizes model configuration files from the cloud collaboration unit, and updates the model parameters of the model storage pool. The real-time inference module, based on the standardized equipment data output by the data access layer and the associated equipment identifiers and operating condition information, matches the corresponding AI diagnostic model from the model storage pool, completes fault diagnosis inference locally, and outputs the fault type, level and confidence level. The logging module fully records the model call time, device identifier, input data characteristics, diagnostic results and inference time, forming a structured call log, which is stored in the data storage layer; the training update module supports four update modes: online training, online update, offline manual update, and model parameter visualization tuning.

[0026] The four update modes are as follows: Online training mode: Receives training instructions for a specified date from the user via the terminal, extracts standardized device data for the corresponding date from the data storage layer or temporary storage area, and encrypts and sends the data back to the cloud collaborative unit via API, whereby the cloud collaborative unit completes incremental model training; Online update mode: After the cloud-based collaborative unit completes model training, it generates an optimized model parameter package and pushes it to the edge diagnostic unit. The training update module automatically verifies the integrity and compatibility of the model parameter package. After the verification is passed, it automatically replaces the parameters of the old model and completes the online model update. The whole process does not require downtime and does not affect the normal operation of diagnostic services. Offline manual update mode: Supports importing offline model parameter packages. Users can update the model through the local interactive interface of the edge diagnostic unit. The training update module completes the manual update after verifying that the offline model parameter package is correct. Model parameter visualization tuning mode: Provides a visualization parameter tuning interface, allowing users to adjust the core parameters of the model according to the real-time operating conditions of the equipment. The tuned parameters take effect in real time, and the tuning effect can be quickly verified through diagnostic results feedback.

[0027] Report generation layer It includes a knowledge base module, a report generation module, and a format conversion module; the knowledge base module has a built-in local knowledge base, which is used to store the mapping relationship between industrial equipment fault types and maintenance measures; The report generation module automatically matches the corresponding entries in the local knowledge base based on the output of the model diagnosis layer, and generates a structured diagnosis report by combining the device identifier, diagnosis time, and operating parameters. The report is divided into a simplified version and a detailed version. The simplified version contains core fault information and emergency handling suggestions for rapid early warning. The detailed version contains equipment operation data statistics, fault source analysis, and maintenance cycle suggestions for operation and maintenance review. The format conversion module supports converting reports to JSON and PDF formats. The simplified version of the report is stored in JSON format by default to meet the needs of real-time push, while the detailed version of the report can be customized to select the storage format.

[0028] Protocol push layer The protocol push layer includes a push strategy module, a protocol adaptation module, and a status feedback module; it supports selecting push strategies and communication protocols according to fault levels, pushes diagnostic reports to the terminal, and has automatic retry and status recording functions for push failures.

[0029] Edge diagnostic unit hardware The edge diagnostic unit is an industrial-grade diagnostic box. In one embodiment of the present invention, its hardware parameters are as follows: Package dimensions: 180mm (L) × 130mm (W) × 40mm (H); Processor: ARM Cortex-A55 quad-core, clock speed 1992MHz; Memory: 4GB DDR4; Storage: 8GB eMMC + M.2 / SATA expansion (supports up to 2TB SSD); Network interfaces: 2 x Gigabit Ethernet, 2 x RS485 (COM3 / 4, COM5 / 6), 2 x CAN (CAN1 / 2), supports Wi-Fi / 4G MPCIE modules; Sensor interface: 8 analog inputs (AI), 16 digital inputs (DI), 8 digital outputs (DO), expanded via J5 header; This unit supports the connection of the following industrial sensors: Vibration sensors: temperature vibration sensors, piezoelectric accelerometers, magnetoelectric velocity sensors; Temperature-related devices: PT100 platinum resistance thermometer, K-type thermocouple, infrared temperature sensor; Electrical equipment: Hall effect current sensors, voltage transformers, power quality analyzers; Process-related: Piezoresistive pressure sensors, strain gauge tension sensors, flow sensors; Status-based: Incremental encoders, proximity switches, displacement sensors.

[0030] Cloud Collaboration Unit The cloud-based collaborative unit includes a model configuration module and an incremental training module. The model configuration module manages AI diagnostic models and parameter configuration files for different industrial equipment under different operating conditions, supports model version management, and records the content and time of each model parameter update for easy backtracking. The incremental training module is used to receive data from the edge diagnostic unit for a specified date, combine it with historical training data in the cloud, and use the incremental training algorithm to train the basic model to generate an optimized model parameter package.

[0031] See Figure 2 A lightweight industrial equipment fault diagnosis method based on IoT and AI, applied to the above system, includes the following steps: Step 1: The edge diagnostic unit adaptively accesses industrial equipment operation data through multiple protocols and converts it into standardized equipment data; Step 2: Store standardized equipment data in a local lightweight database, automatically clean up raw data that exceeds the preset retention time, and mark temporarily stored training data; Step 3: Synchronize the cloud-based collaborative unit or call the local model parameters, match the lightweight AI model to complete local real-time inference, output diagnostic results and record the call log; Step 4: Based on the diagnostic results, match the local knowledge base to generate a tiered diagnostic report and complete the format conversion; Step 5: Select the protocol and policy according to the fault level, push the report to the target terminal and monitor the status; Step 6: Update the model via cloud-based collaborative units online, locally offline, or in parameter tuning mode; Step 7: Iterate collaboratively between edge and cloud, or manually update the model and local knowledge base locally.

[0032] The following is a specific implementation case using fault diagnosis of coating machines in the new energy industry as an application scenario, to further introduce the invention: I. System Deployment The system adopts an architecture of "edge diagnostic unit + optional cloud collaboration unit". The edge diagnostic unit uses an industrial-grade diagnostic box and is deployed in the control cabinet of the coating machine. It can independently complete local closed-loop diagnosis. The cloud collaboration unit is deployed on the enterprise's private cloud server for batch management of multiple diagnostic units and incremental training of models.

[0033] The data acquisition covers temperature sensors in the coating machine oven, current / voltage sensors in the fan, temperature and vibration sensors, tension / speed sensors, and PLC system. It connects to Modbus TCP / RTU and RS485 protocols to achieve unified acquisition of multi-source data.

[0034] II. System Operation Flow Step 1. Data Access and Standardization The protocol adaptation module automatically identifies the communication protocol between the sensor and the PLC, and the format conversion module converts heterogeneous data into standardized device data of "timestamp + device identifier + data type + value + unit", filtering out null values ​​and transient interference data, achieving a data acquisition success rate of 99.68%.

[0035] Step 2. Data Storage and Management The local database uses an embedded SQLite database to store standardized device data and diagnostic logs for the day. The storage strategy module automatically cleans up the original data from 7 days ago every day at midnight, retains the diagnostic records of the past 7 days, and marks the training data to be temporarily stored in the temporary storage area. Specifically, according to a preset 10-second time window (which can be modified according to actual needs via the interface), standardized time-series data from different sources are time-aligned and merged; at the same time, outliers caused by instantaneous interference from sensors are filtered out, completing data cleaning, and the scattered multi-source data are aggregated along the time dimension to form a complete "time-data" set. When the production line is running at full load (coating speed 60m / min, oven temperature 180℃), this step takes an average of ≤50ms.

[0036] Step 3. Model Diagnosis and Inference The model management module loads a lightweight model specific to the coating machine in PKL format. The real-time inference module completes local diagnosis based on standardized data, with an average inference time of 385ms, and outputs the fault type, level and confidence level. The log recording module fully stores the diagnostic process data, which is convenient for fault tracing. Example: Input data: Oven A temperature is 180℃, higher than the set threshold; Fan A current fluctuation exceeds 15%. The output diagnostic results are as follows: the fault type is "aging of heating tube A in oven + wear of bearing A in fan", the fault level is medium, the confidence level is 92%, and the entire inference process has a delay of ≤500ms to ensure timely fault detection.

[0037] Step 4. Report Generation and Format Conversion Step 4.1: The report generation module matches the local coating machine fault knowledge base; For example: Fault type: Oven temperature is too high Diagnosis: The heating element is aging or the temperature controller is malfunctioning. Emergency response plan: Immediately reduce heating power, and replace the heating element after shutdown. Step 4.2: Generate a simple early warning report in JSON format and a detailed review report in PDF format, including fault conclusions, handling suggestions, and running data curves; For example: a simplified report (JSON format): includes information such as device identifier, diagnosis time, confidence level, diagnosis conclusion, and treatment recommendations. Detailed report (PDF format): Includes oven temperature change curves over the past 24 hours, fan current fluctuation statistics, detailed diagnostic output results, etc., for maintenance review.

[0038] Step 5. Multi-protocol push and status monitoring The protocol push layer uses protocols such as HTTP, MQTT, and Webhook to push reports to the MES system and the terminals of operation and maintenance personnel (i.e., third-party systems). If the push fails, it will automatically retry and record the push status.

[0039] Step 6. Model Update and System Iteration When configuring the cloud, the edge diagnostic unit synchronizes the diagnostic data of the day to the cloud every morning at midnight. After the cloud completes incremental training, it pushes the parameter package, and the edge automatically updates the model. At the same time, the cloud updates the coating machine fault knowledge base, and the edge synchronously updates the local knowledge base.

[0040] Without cloud configuration, users import new model packages and knowledge base update files through the local interactive interface of the diagnostic box every quarter, analyze fault patterns by combining local diagnostic logs, adjust model parameters, and complete system optimization.

[0041] III. System Performance Verification This system has been running continuously for over 200 days on the coating machine production line of a new energy materials plant, with a diagnostic accuracy rate of 96.7%. The average CPU load of the system is ≤35%, the memory usage is stable at 1.8GB, and there are no downtimes or data loss, meeting the high stability and high real-time requirements of industrial sites.

[0042] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A lightweight industrial equipment fault diagnosis system based on IoT and AI, characterized in that, It includes an edge diagnostic unit and an optional cloud collaboration unit. The edge diagnostic unit adopts a top-down five-layer architecture, namely a data access layer, a data storage layer, a model diagnostic layer, a report generation layer, and a protocol push layer. Each layer realizes data flow and functional collaboration through standardized interfaces to complete the local full-process closed-loop diagnosis. The cloud collaboration unit and the edge diagnostic unit collaborate asynchronously through a lightweight communication protocol to undertake the tasks of incremental model training and batch device management.

2. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The data access layer includes a protocol adaptation module and a format conversion module; The protocol adaptation module pre-integrates MQTT, HTTP, Modbus TCP / RTU, OPC UA, and RS485 / 232 communication protocols, and automatically identifies the access protocol through data feature identifier bits; The format conversion module converts multi-source heterogeneous raw data into standardized device data and filters out invalid data. The standardized equipment data includes timestamps, equipment identifiers, data types, numerical values, and units. The invalid data includes null values ​​and garbled characters.

3. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The data storage layer includes a local database module and a storage strategy management module; The local database module uses a lightweight embedded database to store standardized device data, diagnostic results, and model call logs. The storage strategy management module executes daily storage and automatic cleanup rules, automatically deleting raw data that has reached the preset retention period at midnight every day, and only retaining diagnostic logs and model records that have not exceeded the preset retention period. It also supports specifying training data for temporary storage.

4. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The model diagnostic layer includes a model management module, a real-time inference module, a log recording module, and a training update module; The model management module has a built-in model storage pool, which stores models in PKL format, periodically synchronizes model configuration files from the cloud collaboration unit, and updates the model parameters of the model storage pool. The real-time inference module, based on the standardized equipment data output by the data access layer and the associated equipment identifiers and operating condition information, matches the corresponding AI diagnostic model from the model storage pool, completes fault diagnosis inference locally, and outputs the fault type, level and confidence level. The log recording module fully records the model call time, device identifier, input data characteristics, diagnostic results and inference time, forming a structured call log, which is stored in the data storage layer; The training update module supports four update modes: online training, online update, offline manual update, and model parameter visualization tuning.

5. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 4, characterized in that, The four update modes of the training update module are as follows: Online training mode: Receives training instructions for a specified date from the user via the terminal, extracts standardized device data for the corresponding date from the data storage layer or temporary storage area, and encrypts and sends the data back to the cloud collaborative unit via API, whereby the cloud collaborative unit completes incremental model training; Online update mode: After the cloud-based collaborative unit completes model training, it generates an optimized model parameter package and pushes it to the edge diagnostic unit. The training update module automatically verifies the integrity and compatibility of the model parameter package. After the verification is passed, it automatically replaces the parameters of the old model and completes the online model update. The whole process does not require downtime and does not affect the normal operation of diagnostic services. Offline manual update mode: Supports importing offline model parameter packages. Users can update the model through the local interactive interface of the edge diagnostic unit. The training update module completes the manual update after verifying that the offline model parameter package is correct. Model parameter visualization tuning mode: Provides a visualization parameter tuning interface, allowing users to adjust the core parameters of the model according to the real-time operating conditions of the equipment. The tuned parameters take effect in real time, and the tuning effect can be quickly verified through diagnostic results feedback.

6. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The report generation layer includes a knowledge base module, a report generation module, and a format conversion module; The knowledge base module has a built-in local knowledge base, which is used to store the mapping relationship between industrial equipment fault types and maintenance measures. The report generation module automatically matches the corresponding entries in the local knowledge base based on the output of the model diagnosis layer, and generates a structured diagnosis report by combining the device identifier, diagnosis time, and operating parameters. The report is divided into a simplified version and a detailed version. The simplified version contains core fault information and emergency handling suggestions for rapid early warning. The detailed version contains equipment operation data statistics, fault source analysis, and maintenance cycle suggestions for operation and maintenance review. The format conversion module supports converting reports to JSON and PDF formats. The simplified version of the report is stored in JSON format by default to meet the needs of real-time push, while the detailed version of the report can be customized to select the storage format.

7. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The protocol push layer includes a push strategy module, a protocol adaptation module, and a status feedback module; it supports selecting push strategies and communication protocols according to fault levels, pushes diagnostic reports to the terminal, and has automatic retry and status recording functions for push failures.

8. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The edge diagnostic unit is an industrial-grade diagnostic box with the following hardware configuration: ARM Cortex-A55 quad-core processor, 4GB DDR4 memory, 8GB eMMC storage, and support for M.2 / SATA expansion; it has Gigabit Ethernet, RS485, CAN, Wi-Fi / 4G network interfaces, as well as multiple analog and digital sensor interfaces, which can be connected to vibration, temperature, electrical, process, and status industrial sensors.

9. The lightweight industrial equipment fault diagnosis system based on IoT and AI according to claim 1, characterized in that, The cloud-based collaborative unit includes a model configuration module and an incremental training module; The model configuration module manages AI diagnostic models and parameter configuration files for different industrial equipment under different operating conditions, supports model version management, and records the content and time of each model parameter update. The incremental training module is used to receive data from the edge diagnostic unit for a specified date, combine it with historical training data in the cloud, and use the incremental training algorithm to train the basic model to generate an optimized model parameter package.

10. A method for fault diagnosis of lightweight industrial equipment based on the Internet of Things and AI, characterized in that, Applied to the system according to any one of claims 1-9, comprising the following steps: Step 1: The edge diagnostic unit adaptively accesses industrial equipment operation data through multiple protocols and converts it into standardized equipment data; Step 2: Store standardized equipment data in a local lightweight database, automatically clean up raw data that exceeds the preset retention time, and mark temporarily stored training data; Step 3: Synchronize the cloud-based collaborative unit or call the local model parameters, match the lightweight AI model to complete local real-time inference, output diagnostic results and record the call log; Step 4: Based on the diagnostic results, match the local knowledge base to generate a tiered diagnostic report and complete the format conversion; Step 5: Select the protocol and policy according to the fault level, push the report to the target terminal and monitor the status; Step 6: Update the model via cloud-based collaborative units online, locally offline, or in parameter tuning mode; Step 7: Iterate collaboratively between edge and cloud, or manually update the model and local knowledge base locally.

Images