Predictive maintenance method and related product

Abstract

The embodiment of the invention provides a predictive maintenance method and a related product, and relates to the technical field of data processing. The real-time data of the target industrial equipment is analyzed by using the edge layer, and the local maintenance decision can be determined and adopted to avoid accidental shutdown of the target industrial equipment by identifying the potential fault of the target industrial equipment in advance. This not only improves the reliability of the target industrial equipment, but also reduces unplanned time delay and maintenance costs. Or, the central layer is utilized to analyze the processed real-time data of all the industrial equipment in the production system, and the cooperative work condition of all the industrial equipment of the whole production system is considered, so that a deeper potential fault can be found, and a more comprehensive global maintenance decision is determined and adopted; the invention aims to maximize the overall productivity of a production system and minimize downtime.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and in particular, to a predictive maintenance method and related products. Background Art

[0002] In modern industrial production, the efficient and stable operation of industrial equipment (such as compressors, motors, conveyor belts, etc.) is the basis for maintaining the continuity of the production process and the economic benefits of enterprises. Therefore, ensuring that industrial equipment is in the best working condition and reducing or avoiding the failure of industrial equipment have become important means to improve production efficiency, reduce operating costs, and enhance the competitiveness of enterprises.

[0003] In the related art, the maintenance decision for industrial equipment includes corrective maintenance (also known as post-failure maintenance). Corrective maintenance refers to starting maintenance when an industrial equipment fails and affects production. The advantage of corrective maintenance is that it is simple to implement and does not require a large amount of resources to be invested in inspection or prediction work in advance, so the initial cost is relatively low.

[0004] However, since corrective maintenance is carried out only after a failure has occurred, this will inevitably lead to the suspension of industrial production, resulting in time delays and economic losses. Summary of the Invention

[0005] To solve the above problems, the embodiments of this application provide a predictive maintenance method and related products, which not only improve the reliability of industrial equipment, but also reduce unplanned time delays and maintenance costs.

[0006] In a first aspect, this application discloses a predictive maintenance method, which is applied to a predictive maintenance system. The predictive maintenance system includes a device layer, an edge layer, and a central layer. The device layer is communicatively connected to the edge layer, and the edge layer is communicatively connected to the central layer. The method includes: the device layer obtains real-time data of a target industrial equipment in a production system; the device layer inputs the real-time data of the target industrial equipment into the edge layer, so that the edge layer generates a local maintenance decision for the target industrial equipment; the edge layer performs maintenance on the target industrial equipment according to the local maintenance decision; or, the device layer inputs the real-time data of all industrial equipment in the production system into the edge layer, so that the edge layer processes the real-time data; the edge layer inputs the processed real-time data into the central layer, so that the central layer generates a global maintenance decision for all industrial equipment in the production system; the central layer performs maintenance on all industrial equipment in the production system according to the global maintenance decision.

[0007] It can be seen from this that the present application analyzes the real-time data of the target industrial equipment by using the edge layer, and can identify potential faults of the target industrial equipment in advance, determine and take local maintenance decisions to avoid unexpected shutdowns of the target industrial equipment. This not only improves the reliability of the target industrial equipment, but also reduces unplanned time delays and maintenance costs. Alternatively, the present application analyzes the processed real-time data of all industrial equipment in the production system by using the central layer. Considering the collaborative working conditions of all industrial equipment in the entire production system, deeper potential faults can be discovered, and then more comprehensive global maintenance decisions can be determined and taken to maximize the overall productivity of the production system and minimize the downtime.

[0008] In some specific implementation manners, the target industrial equipment is any one industrial equipment in the production system, or multiple industrial equipment located on the same production line in the production system, or multiple industrial equipment in the production system with a similarity of operating states higher than a similarity threshold, where the operating states include any one or more of temperature, pressure, vibration frequency, current, and voltage.

[0009] In some specific implementation manners, the edge layer includes a first decision generation model; inputting the real-time data of the target industrial equipment into the edge layer to enable the edge layer to generate a local maintenance decision for the target industrial equipment includes: inputting the real-time data of the target industrial equipment into the first decision generation model to enable the first decision generation model to generate a local maintenance decision for the target industrial equipment.

[0010] In some specific implementation manners, the first decision generation model is trained in the following manner: obtaining the historical operation data of the target industrial equipment and the first historical fault that occurred when the target industrial equipment was operating under the historical operation data; determining, based on the first historical fault, the historical local maintenance decision for the target industrial equipment to solve the first historical fault; and training a machine learning model according to the historical operation data and the historical local maintenance decision to obtain the first decision generation model.

[0011] In some specific implementation manners, the first decision generation model includes a first sub-model and a second sub-model; inputting the real-time data of the target industrial equipment into the edge layer to enable the edge layer to generate a local maintenance decision for the target industrial equipment includes: inputting the real-time data of the target industrial equipment into the first sub-model to obtain prediction fault information corresponding to the real-time data of the target industrial equipment; and inputting the prediction fault information into the second sub-model to obtain a local maintenance decision for the target industrial equipment corresponding to the prediction fault information.

[0012] In some specific implementation manners, after obtaining the prediction fault information corresponding to the real-time data of the target industrial equipment, the method further includes: triggering an alarm indication when the prediction fault information indicates that the fault degree of the target industrial equipment is higher than a preset degree threshold.

[0013] In some specific implementation manners, the central layer includes a second decision-making generation model; the processed real-time data is input into the central layer so that the central layer generates global maintenance decisions for all industrial devices in the production system, including: inputting the processed real-time data into the second decision-making generation model so that the second decision-making generation model generates global maintenance decisions for all industrial devices in the production system.

[0014] In some specific implementation manners, the second decision-making generation model is trained as follows: obtaining historical operation data of all industrial devices in the production system, and second historical faults that all industrial devices in the production system have during the operation under the historical operation data; determining historical global maintenance decisions of the production system for solving the second historical faults based on the second historical faults; training a machine learning model according to the historical operation data, the historical global maintenance decisions, and the influence characteristics among all industrial devices in the production system to obtain the second decision-making generation model.

[0015] In some specific implementation manners, after obtaining the real-time data of the target industrial device in the production system, the method further includes: if the real-time data of the target industrial device does not meet the preset conditions, triggering an alarm indication, where the preset conditions include any one or more of the real-time temperature data of the industrial device being lower than the preset temperature threshold, the growth rate of the real-time temperature data of the industrial device being lower than the first preset growth threshold, the real-time pressure data of the industrial device being lower than the maximum safety threshold and higher than the minimum safety threshold, the growth rate of the real-time pressure data of the industrial device being lower than the second preset growth threshold, the decrease rate of the real-time pressure data of the industrial device being lower than the preset decrease threshold, the real-time humidity data of the industrial device being lower than the preset humidity threshold, the growth rate of the real-time humidity data of the industrial device being lower than the third preset growth threshold, and the real-time flow data of the industrial device being higher than the preset flow threshold.

[0016] In some specific implementation manners, the edge layer processes the real-time data, including: the edge layer performs format-unifying processing on the real-time data of all industrial devices in the production system.

[0017] In a second aspect, the present application discloses an electronic device, which includes: a memory for storing a computer program or computer instructions; a processor for executing the computer program or computer instructions stored in the memory, so that the electronic device executes the steps performed by the device layer, the edge layer, or the central layer in the predictive maintenance method as in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of a predictive maintenance system provided by an embodiment of the present application;

[0019] Figure 2A signaling diagram of a predictive maintenance system provided by an embodiment of the present application;

[0020] Figure 3 Another signaling diagram of a predictive maintenance system provided by an embodiment of the present application;

[0021] Figure 4 A training flow chart of a first decision-making generation model provided by an embodiment of the present application;

[0022] Figure 5 A training flow chart of a second decision-making generation model provided by an embodiment of the present application;

[0023] Figure 6 A schematic diagram of another predictive maintenance system provided by an embodiment of the present application;

[0024] Figure 7 A flow chart of a predictive maintenance method provided by an embodiment of the present application. Detailed implementation manners

[0025] First, the technical terms involved in the embodiments of the present application will be explained:

[0026] EtherCAT (Ethernet for Control Automation Technology) is a communication protocol for real-time Ethernet control systems, with the characteristics of high real-time performance (able to achieve microsecond-level response time, suitable for application scenarios requiring high-precision synchronization), efficient data transmission using Ethernet frames, reducing latency, and flexibility (supporting various topologies such as linear, star, and tree, facilitating system expansion).

[0027] OPC UA (Open Platform Communications Unified Architecture) is a platform-independent, service-oriented architecture for securely and reliably exchanging real-time data, with the characteristics of platform independence (able to run on different operating systems and hardware platforms) and security (providing authentication, authorization, and encryption functions).

[0028] Profinet (Process FieldNetwork) is an open standard based on industrial Ethernet for real-time communication in factory automation environments, with the characteristics of real-time communication (supporting hard real-time and soft real-time communication to meet the needs of different application scenarios) and reliability (able to operate stably in harsh industrial environments).

[0029] The terms "first" and "second" in the embodiments of the present application are only for descriptive purposes, and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features.

[0030] The embodiments of the present application provide a predictive maintenance method and related products. The present application analyzes the real-time data of target industrial equipment using the edge layer, and can identify potential faults of the target industrial equipment in advance, determine and take local maintenance decisions to avoid unexpected downtime of the target industrial equipment. This not only improves the reliability of the target industrial equipment, but also reduces unplanned time delays and maintenance costs. Alternatively, the present application analyzes the processed real-time data of all industrial equipment in the production system using the central layer. Considering the collaborative working conditions of all industrial equipment in the entire production system, deeper potential faults can be discovered, and then more comprehensive global maintenance decisions can be determined and taken to maximize the overall productivity of the production system and minimize downtime.

[0031] See Figure 1 , which is a schematic diagram of a predictive maintenance system provided by an embodiment of the present application. As Figure 1 can be seen, the predictive maintenance system 10 includes a device layer (Device Layer, DL) 11, an edge layer (Edge Layer, EL) 12, and a central layer (Central Layer, CL) 13.

[0032] First, the device layer 11 will be explained: The device layer 11 is at the bottom layer of the predictive maintenance system 10, and includes sensors and embedded processors installed on industrial equipment. Among them, the sensors are used to collect real-time data of industrial equipment (such as real-time temperature data, real-time pressure data, and real-time vibration data, etc.). The embedded processor is not only used to obtain the real-time data collected by the sensors, but also used to process the real-time data and transfer it to the edge layer 12. Specifically, the embedded processor can filter out abnormal data in the real-time data (that is, real-time data that significantly deviates from the normal range or does not conform to logic due to sensor failures, environmental interference, or other external factors), and can also perform signal preprocessing (including denoising, normalization, and data smoothing, etc.) operations to ensure the quality and effectiveness of the real-time data transferred to the edge layer 12.

[0033] Secondly, an explanation of the edge layer 12 is provided: The edge layer 12 is the middle layer between the device layer 11 and the central layer 13 in the predictive maintenance system 10. The edge layer 12 is used to determine corresponding local maintenance decisions based on the real-time data of the target industrial device, and thus perform predictive maintenance on the target industrial device according to the local maintenance decisions. In a specific implementation, the edge layer includes a data processing module, which is used to input the real-time data of the target industrial device into the first decision-making model located in the edge layer 12, so that the first decision-making model outputs a local maintenance decision corresponding to the real-time data of the target industrial device. Thereby, the downtime of the target industrial device can be reduced, and the reliability and production efficiency of the target industrial device can be improved. Alternatively, the edge layer 12 is also used to process the real-time data (such as performing protocol unification processing on the real-time data, removing noise data from the real-time data, and filtering invalid data from the real-time data) after receiving the real-time data of all industrial devices in the production system. Subsequently, the processed real-time data is input into the central layer 13. Among them, the edge layer includes an edge server and an edge gateway. The data processing module and the first decision-making model run on the edge server, and the edge gateway realizes the communication between the edge layer 12 and the central layer 13 and the device layer 11; by adopting the edge server, the edge layer 12 has powerful computing capabilities, the real-time data can be transmitted to the edge layer 12 and stored locally, and the data processing module in the edge layer 12 has a fast data processing speed, which can reduce data transmission latency.

[0034] Finally, an explanation of the central layer 13 is provided: The central layer is the upper layer of the edge layer 12 in the predictive maintenance system 10 and includes a data processing module. The data processing module is used to integrate the processed real-time data of all industrial devices in the production system transmitted from the edge layer 12 (that is, the real-time data after completing protocol unification processing, removing noise data, and filtering invalid data), and input the processed real-time data of all industrial devices in the production system into the second decision-making model located in the central layer 13, so that the second decision-making model uses cloud computing resources and data storage capabilities to output a global maintenance decision corresponding to the processed real-time data of all industrial devices in the production system, and thus perform predictive maintenance on the entire production system according to the global maintenance decision. Thereby, not only the state of a single industrial device can be considered, but also the collaborative working conditions of all industrial devices in the entire production system can be considered, so that deeper potential faults can be discovered, and then more accurate and comprehensive global maintenance decisions can be determined and taken to maximize the overall productivity of the production system and minimize downtime. The central layer may include a server, such as an AI server, etc.

[0035] It can be understood that the predictive maintenance system 10 can improve the data processing efficiency by performing distributed computing in the three - layer architectures of the device layer 11, the edge layer 12, and the central layer 13, avoid the performance bottleneck caused by centralized computing, and achieve efficient predictive maintenance.

[0036] See Figure 2 , which is a signaling diagram of a predictive maintenance system provided by an embodiment of this application.

[0037] S201: The device layer acquires real - time data of industrial devices in the production system.

[0038] The real - time data may include at least one of real - time temperature data, real - time pressure data, real - time vibration data, real - time humidity data, and real - time flow data.

[0039] In some specific implementation manners, real - time temperature data can be collected by a temperature sensor. For example, the temperature sensor can be deployed at positions such as the motor, bearing, and hydraulic system of the industrial device to monitor whether the industrial device has abnormal temperature rise due to overload, friction, etc., to ensure the normal operation of the industrial device.

[0040] In some other specific implementation manners, real - time pressure data can be collected by a pressure sensor. For example, the pressure sensor can be deployed inside the pipeline of the industrial device to detect the pressure change of the liquid or gas inside the pipeline, thereby helping to identify possible problems such as blockage and leakage to ensure the normal operation of the industrial device.

[0041] In some other specific implementation manners, real - time vibration data can be collected by a vibration sensor. For example, the vibration sensor can be deployed on industrial devices that generate vibrations during operation (such as motors, fans, pumps, etc.) to capture the minute vibrations generated during the operation of the industrial device, and thus discover early fault signs such as imbalance, looseness, and wear through spectrum analysis, and give early warnings of potential problems to ensure the normal operation of the industrial device.

[0042] In some other specific implementation manners, real - time humidity data can be collected by a humidity sensor. For example, the humidity sensor can be installed inside the electrical cabinet that is easily affected by moisture to monitor the environmental humidity level inside the electrical cabinet, thereby preventing electronic components from being corroded or damaged due to excessive humidity to ensure the normal operation of the industrial device.

[0043] In some other specific implementation manners, real - time flow data can be collected by a flow sensor. For example, the real - time flow data can be deployed inside the pipeline of the industrial device to detect the flow rate of the passing liquid or gas to ensure that the flow rate meets the expectations and avoid problems caused by insufficient or excessive flow rate to ensure the normal operation of the industrial device.

[0044] It should be noted that, in order to ensure the accuracy and timeliness of real-time data, the above sensors can set different sampling rates according to the change rate of the monitored parameters (such as temperature, pressure, vibration, etc.). Exemplarily, parameters that change rapidly (such as vibration) may require a higher sampling rate, while relatively stable parameters (such as temperature) can use a lower sampling rate. The specific sampling rate should be determined by the actual application situation, and this application does not make any limitations in this regard.

[0045] It should also be noted that when deploying sensors, considering the harsh conditions in the industrial environment (such as high temperature, humidity, corrosion, etc.), the design of the sensors also needs to have good durability and anti-interference capabilities to ensure long-term stable data acquisition performance. The specific sensors should be determined by the actual application situation, and this application does not make any limitations in this regard.

[0046] It should further be noted that after the device layer obtains the real-time data of industrial devices in the production system, it is also necessary to filter out abnormal data in the real-time data and perform signal preprocessing operations. Abnormal data refers to those real-time data that significantly deviate from the normal range or do not conform to logic, and these abnormalities may be caused by sensor failures, environmental interference, or other external factors. Specifically, the upper and lower limits of the real-time data can be set, and the real-time data outside this range is regarded as abnormal data and filtered out. Exemplarily, if the normal operating temperature range of a temperature sensor is from 0°C to 100°C, then any data below 0°C or above 100°C will be regarded as abnormal data. Thereby, the overall quality of the real-time data can be improved to ensure the accuracy of local maintenance decisions and global maintenance decisions. Signal preprocessing refers to a series of operations performed on the original signal to improve the signal quality and make it more suitable for subsequent analysis or transmission. Specifically, signal preprocessing can include denoising (i.e., removing noise through a filter to make the signal clearer), normalization (i.e., adjusting real-time data of different magnitudes or units to the same scale for easy comparison and analysis), data smoothing (i.e., applying smoothing techniques to reduce the fluctuations of real-time data in a short period of time to make the trend more obvious), etc. This application does not make any limitations on the specific signal preprocessing methods.

[0047] S202: The device layer transfers the real-time data of the target industrial device to the edge layer.

[0048] Specifically, the device layer transfers the real-time data of the target industrial device to the edge layer through industrial protocols (such as EtherCAT, OPC UA, Profinet, etc.).

[0049] It should be noted that in the case of limited bandwidth, in order to reduce the transmission load, the device layer can adopt an appropriate compression algorithm to compress the real-time data to reduce the data volume, and then transfer the compressed real-time data to the edge layer. In some specific implementation manners, the compression methods include lossless compression and feature extraction and dimensionality reduction compression. Among them, lossless compression refers to using algorithms (such as LZ77, Huffman coding) to reduce the volume of real-time data without losing any information, which is applicable to occasions where data integrity needs to be maintained. Feature extraction and dimensionality reduction compression refers to extracting the most representative features in the real-time data through methods such as principal component analysis and linear discriminant analysis, so as to retain the core information while reducing the data volume. The present application does not limit the specific compression method.

[0050] It should also be noted that if an error occurs (such as packet loss) during the process of transferring the real-time data from the device layer to the edge layer, the corresponding error recovery strategy can be executed according to the definition of the above industrial protocol, such as automatically requesting retransmission of the lost real-time data. The present application does not limit the specific error recovery strategy.

[0051] S203: The edge layer generates a local maintenance decision for the target industrial device based on the real-time data of the target industrial device, so that the edge layer can perform predictive maintenance on the target industrial device according to the local maintenance decision.

[0052] In some specific implementation manners, the edge layer includes a first decision generation model. The first decision generation model is designed for the target industrial device and is used to determine whether predictive maintenance is required and determine the optimal predictive maintenance decision by evaluating the real-time data of the target industrial device. Specifically, the edge layer inputs the real-time data of the target industrial device into the first decision generation model located in the edge layer, so that the first decision generation model evaluates the real-time data of the target industrial device and outputs the corresponding local maintenance decision. In one example, if the first decision generation model predicts that a certain component of the target industrial device is about to fail, it will output a local maintenance decision suggesting to replace the component in advance. In another example, if the first decision generation model finds that some indicators of the target industrial device are abnormal but not yet serious enough to constitute an emergency, it will output a local maintenance decision to arrange regular inspections.

[0053] It should be noted that the target industrial device can refer to any industrial device in the production system, or the target industrial device can also refer to multiple industrial devices on the same production line in the production system, or the target industrial device can also refer to multiple industrial devices in the production system with a similarity of operating states (such as temperature, pressure, vibration frequency, current intensity, voltage level, etc.) higher than a similarity threshold. Among them, high similarity means that the operating states of these industrial devices are very close within the same time period, and they may face the same risks or exhibit similar failure modes.

[0054] It should also be noted that the above embodiments are explained by training the first decision generation model. In practical applications, the first fault judgment model can also be trained. By inputting the real-time data of the target industrial device into the first fault judgment model, the first fault judgment model can output the corresponding fault cause and fault time. Subsequently, the corresponding local maintenance decision is determined according to the fault cause and fault time. This application does not make any limitations on this.

[0055] It should also be noted that the edge layer also has a local storage function for temporarily storing the real-time data received from the device layer to ensure the integrity and traceability of the real-time data.

[0056] After obtaining the local maintenance decision provided by the first decision generation model, the edge layer will perform predictive maintenance on the target industrial device according to the local maintenance decision. Exemplarily, the edge layer can trigger a prompt notification to notify the on-site technicians to prepare the necessary tools and spare parts, and then execute the maintenance task according to the plan of the local maintenance decision.

[0057] S204: The edge layer processes the real-time data of all industrial devices in the production system and transfers the processed real-time data to the central layer.

[0058] After obtaining the real-time data of all industrial devices in the production system, the edge layer also needs to unify the format of the real-time data of all industrial devices in the production system, and eliminate the noise data in the real-time data of all industrial devices in the production system, and filter the invalid data in the real-time data of all industrial devices in the production system. Among them, eliminating the noise data in the real-time data of all industrial devices in the production system means filtering the noise data in the real-time data through a filter (such as a low-pass filter, a high-pass filter) to make the signal clearer. Filtering the invalid data in the real-time data of all industrial devices in the production system means identifying and removing the real-time data that significantly deviates from the normal range or does not conform to logic. Subsequently, the edge layer transfers the processed real-time data to the central layer.

[0059] S205: The central layer generates global maintenance decisions for all industrial devices in the production system based on the processed real-time data, so that the central layer performs predictive maintenance on all industrial devices in the production system according to the global maintenance decisions.

[0060] In some specific implementation manners, the central layer includes a second decision-making generation model. The second decision-making generation model is obtained by training a machine learning model based on the historical operation data of all industrial devices in the production system, the historical global maintenance decisions, and the influence characteristics among all industrial devices, and can model and predict the long-term operation trends and potential problems of all industrial devices in the production system. The machine learning model extracts complex features from the processed real-time data of all industrial devices in the production system through a multi-layer neural network structure, and combines time series analysis to capture the periodic changes and abnormal patterns in the operation of industrial devices. Through the analysis of these features, the machine learning model can output more accurate and comprehensive global maintenance decisions. Thus, the second decision-making generation model located in the central layer not only considers the real-time data of a single industrial device, but also pays attention to the collaborative work and mutual influence of all industrial devices in the entire production system, such as the physical connections, data transmission relationships, and logical associations among all industrial devices in the production system (for example, the change in the rotation speed of a certain motor may affect the working efficiency of the pump connected to it), so as to obtain more accurate and comprehensive global maintenance decisions.

[0061] After obtaining the global maintenance decisions provided by the second decision-making generation model, the central layer performs predictive maintenance on the entire production system according to the global maintenance decisions. This can not only early warn of the failures of all industrial devices in the production system, but also optimize the operation efficiency of all industrial devices in the production system, extend the service life of all industrial devices in the production system, and thus significantly improve the overall reliability and production efficiency of the production system.

[0062] It should be noted that in practical applications, the choice between global maintenance and local maintenance depends on the following factors: First, if the interdependence among industrial devices in the production system is complex and the failure of one industrial device may affect other industrial devices, global maintenance should be selected; on the contrary, if the industrial devices are highly independent, local maintenance can be selected. Second, during the production peak period, it is necessary to minimize the downtime and local maintenance should be given priority; while in the off-season of production or during the regular maintenance period, global maintenance can be arranged. Third, according to the available human, material resources and budget situation, decide whether to carry out comprehensive maintenance or local maintenance. The application does not limit the factors for specifically choosing global maintenance or local maintenance.

[0063] In summary, the present application discloses a predictive maintenance system, which includes an equipment layer, an edge layer, and a central layer. The present application uses the first decision-making generation model located in the edge layer to analyze the real-time data of the target industrial equipment, and can identify potential faults of the target industrial equipment in advance, determine and take local maintenance decisions to avoid unexpected downtime of the target industrial equipment. This not only improves the reliability of the target industrial equipment, but also reduces unplanned time delays and maintenance costs. Alternatively, the present application uses the second decision-making generation model located in the central layer to analyze the processed real-time data of all industrial equipment in the production system. Considering the collaborative working conditions of all industrial equipment in the entire production system, deeper potential faults can be discovered, and then more comprehensive global maintenance decisions can be determined and taken to maximize the overall productivity of the production system and minimize downtime.

[0064] See Figure 3 , which is a signaling diagram of another predictive maintenance system provided by an embodiment of the present application.

[0065] S301: Sensors in the equipment layer collect real-time data of industrial equipment in the production system.

[0066] It should be noted that the equipment layer can cover one or more different production lines, and each production line may represent a specific product manufacturing process or process step. Usually, several industrial equipment (such as robots, conveyor belts, etc.) are equipped on each production line. Multiple sensors, such as temperature sensors, pressure sensors, vibration sensors, and current sensors, etc., are installed inside or outside each industrial equipment to detect the real-time data of the industrial equipment. Regarding the specific number of production lines, industrial equipment, and sensors, the present application does not make any limitations.

[0067] S302: When the real-time data of the industrial equipment does not meet the preset conditions, the embedded processor in the equipment layer triggers an alarm indication.

[0068] The real-time data of the industrial equipment can be used for simple fault warning. Exemplarily, the preset conditions for fault warning include any one or more of the real-time temperature data of the industrial equipment being lower than the preset temperature threshold (or the growth rate of the real-time temperature data being lower than the first preset growth threshold), the real-time pressure data being lower than the maximum safety threshold and higher than the minimum safety threshold (or the growth rate of the real-time pressure data being lower than the second preset growth threshold, or the reduction rate of the real-time pressure data being lower than the preset reduction threshold), the real-time humidity data being lower than the preset humidity threshold (or the growth rate of the real-time humidity data being lower than the third preset growth threshold), and the real-time flow data being higher than the preset flow threshold.

[0069] In a specific implementation, if the embedded processor determines that the real-time temperature data is higher than or equal to the preset temperature threshold (or the growth rate of the real-time temperature data is higher than or equal to the first preset growth threshold), an alarm indication is triggered to prompt that there may be problems such as industrial equipment overload (specifically referring to industrial equipment with high-load operation such as motors, pumps, compressors, etc.), excessive friction of industrial equipment (specifically referring to industrial equipment relying on mechanical movement such as bearings, gearboxes, conveyor belts, etc.).

[0070] In another specific implementation, if the embedded processor determines that the real-time pressure data is higher than or equal to the set maximum safety threshold (or the growth rate of the real-time pressure data is higher than or equal to the second preset growth threshold), or the real-time pressure data is less than or equal to the set minimum safety threshold (or the decrease rate of the real-time pressure data is higher than or equal to the preset decrease threshold), an alarm indication is triggered to prompt that there may be problems such as industrial equipment blockage (specifically referring to industrial equipment in fluid transmission systems such as pipelines, valves, filters, etc.), industrial equipment leakage (specifically referring to industrial equipment used for connection such as pipelines, joints, seals).

[0071] In another specific implementation, if the embedded processor determines that the real-time vibration data exhibits abnormal spectral characteristics, an alarm indication is triggered to prompt relevant technicians to pay attention to potential imbalance or wear conditions.

[0072] In another specific implementation, if the embedded processor determines that the processed real-time humidity data is higher than or equal to the preset temperature threshold (or the growth rate of the real-time humidity data is higher than or equal to the third preset growth threshold), an alarm indication is triggered to suggest taking dehumidification measures to avoid damage to industrial equipment caused by moisture.

[0073] In another specific implementation, if the embedded processor determines that the processed real-time flow data is lower than the preset flow threshold, an alarm indication is triggered to prompt to ensure timely restoration of normal fluid transmission.

[0074] Thus, by setting the above fault warning function, the self-diagnosis ability of the predictive maintenance system can be enhanced, enabling relevant technicians to identify and solve potential problems at an early stage, thereby effectively preventing the occurrence of major faults and ensuring the stable operation of industrial equipment.

[0075] S303: The embedded processor at the device layer transmits the real-time data of the target industrial equipment to the edge layer.

[0076] In some specific implementations, the target industrial equipment can refer to any industrial equipment in the production system, or the target industrial equipment can also refer to multiple industrial equipment on the same production line in the production system, or the target industrial equipment can also refer to multiple industrial equipment with a similarity of operating states higher than the similarity threshold in the production system.

[0077] It should be noted that in an industrial environment, the real-time data collected by sensors may contain noise or outliers. If this abnormal data is transmitted to the edge layer without being processed, it may lead to incorrect maintenance decisions. Therefore, before the embedded processor in the device layer transmits the real-time data to the edge layer, it can first identify and exclude the abnormal data that significantly exceeds the normal range in the real-time data, and transmit the preprocessed real-time data to the edge layer. Exemplarily, the Z-Score detection method can be used to identify abnormal data. The Z-Score detection method means that if the Z-Score of a certain real-time data exceeds a set threshold (usually ±3), then this real-time data is considered abnormal data. For the specific identification method, this application does not make any limitations. Thus, after identifying and excluding abnormal data, what is obtained is cleaner and more accurate preprocessed real-time data. This step improves the quality of subsequent analysis and reduces the possibility of false alarms.

[0078] S304: The data processing module in the edge layer inputs the real-time data into the first decision-making generation model located in the edge layer to obtain a local maintenance decision corresponding to the real-time data, so that the data processing module in the edge layer can perform predictive maintenance on the target industrial device according to the local maintenance decision.

[0079] The first decision-making generation model is designed for the target industrial device and is used to determine whether predictive maintenance is required and to determine the optimal predictive maintenance decision by evaluating the real-time data of the target industrial device. See Figure 4 , which is a training flow chart of a first decision-making generation model provided by an embodiment of this application. The first decision-making generation model is specifically trained through the following process:

[0080] A1: Obtain the historical operation data of the target industrial device.

[0081] The historical operation data includes at least one of historical temperature data, historical vibration data, historical pressure data, and historical flow data. These historical operation data are collected from the sensors set on the target industrial device to ensure that the machine learning model can learn the operation data of the target industrial device under normal and abnormal conditions.

[0082] It should be noted that for the convenience of subsequent data processing, all historical operation data needs to be first converted into a unified data format to ensure the consistency and accuracy of the historical operation data and improve the quality and efficiency of subsequent decision-making generation.

[0083] A2: Determine the first historical fault of the target industrial device when operating under the historical operation data.

[0084] The first historical failure refers to the failure events and their related details that occur during the operation of the target industrial equipment under historical operation data, including the failure type (such as mechanical wear, electrical short circuit, hydraulic system leakage, etc.), occurrence time, scope of influence (referring to the degree of influence of the failure on the production process, other connected equipment, or the entire production line, which may involve downtime, production loss, safety risks, etc.), failure location, etc.

[0085] A3: Based on the first historical failure, determine the historical local maintenance decision of the target industrial equipment to solve the first historical failure.

[0086] The historical local maintenance decision refers to the specific maintenance decision taken for the target industrial equipment for the first historical failure, such as replacing components, adjusting parameters, regular inspections, etc. Exemplarily, for the first historical failure of temperature increase within a certain period, the historical local maintenance decision may be "increase the coolant flow rate of the target industrial equipment". It can be understood that the historical local maintenance decision is not only to repair the occurred failure, but more importantly, it is the optimal maintenance decision taken to prevent the failure from occurring again.

[0087] A4: Train a machine learning model based on the historical operation data and the historical local maintenance decision to obtain the first decision generation model.

[0088] First, select a suitable machine learning algorithm, such as the Decision Tree algorithm, Random Forest algorithm, Support Vector Machine algorithm, and Neural Network algorithm, etc. Second, use the historical operation data as the input of the machine learning model and the historical local maintenance decision as the output label of the machine learning model, and train the machine learning model through the machine learning algorithm. It can be understood that during the model training process of the machine learning model, the machine learning model will generate a predicted local maintenance decision based on the input historical operation data, compare the predicted local maintenance decision with the historical local maintenance decision, and measure the difference between the predicted local maintenance decision and the historical local maintenance decision by setting a loss function (such as the mean square error MSE or cross-entropy loss function). Subsequently, adopt the gradient descent algorithm to adjust the model parameters of the machine learning model to reduce the difference between the predicted local maintenance decision and the historical local maintenance decision. Through repeated iterative training, gradually improve the accuracy of the predicted local maintenance decision output by the machine learning model until the error between the predicted local maintenance decision generated by the machine learning model and the historical local maintenance decision reaches the minimum (close to 0), and finally obtain a first decision generation model that can automatically recommend local maintenance decisions based on real-time data.

[0089] Once the first decision-making generation model is trained and deployed to the data processing module at the edge layer, the first decision-making generation model can start subsequent processing of real-time data: the real-time data of the target industrial device is input into the first decision-making generation model, and the first decision-making generation model analyzes and evaluates the real-time data of the target industrial device to output corresponding local maintenance decisions.

[0090] It should be noted that the first decision-making generation model can also include a first sub-model and a second sub-model. Among them, the data processing module at the edge layer inputs the real-time data into the first sub-model of the first decision-making generation model to obtain prediction fault information corresponding to the real-time data. Exemplarily, if the real-time data indicates that the vibration frequency of the motor suddenly increases, then the first sub-model may predict that this is an early fault signal caused by bearing wear. Subsequently, the data processing module at the edge layer inputs the prediction fault information into the second sub-model of the first decision-making generation model to obtain local maintenance decisions corresponding to the prediction fault information. Exemplarily, for the above-mentioned bearing wear problem, the second sub-model may recommend replacing the bearing during the next planned downtime and also recommend increasing the frequency of lubricant use. This hierarchical design enables the first decision-making generation model to process real-time data more precisely and ultimately generate more accurate local maintenance decisions, thereby maintaining a high level of production efficiency and safety of the target industrial device.

[0091] Furthermore, if the fault degree of the target industrial device characterized by the prediction fault information determined by the first sub-model of the first decision-making generation model is higher than the preset degree threshold, an alarm indication is triggered. Thus, by setting the preset degree threshold, a warning can be issued before the fault develops to the critical point, enabling relevant technicians to take prompt actions, which helps prevent small problems from evolving into major faults and thus reduces the risk of unexpected downtime of the target industrial device.

[0092] It should also be noted that the data processing module at the edge layer can also save the real-time data to the local storage module at the edge layer. The local storage module can ensure that the real-time data is not lost and supports the generation of local maintenance decisions (i.e., inputting the real-time data in the local storage module into the first decision-making generation model to obtain corresponding local maintenance decisions).

[0093] It should also be noted that the data processing module at the edge layer can be located in the edge server. For this, the present application makes no limitation.

[0094] Thus, when an abnormality occurs in the target industrial device, the edge layer can immediately give a local warning or take necessary local maintenance decisions. And even in the case of unstable or interrupted network, the local storage module at the edge layer can still ensure the integrity of the real-time data, ensuring the continuous operation and data security of the predictive maintenance system.

[0095] S305: The multi - protocol compatibility module in the edge layer processes the real - time data of all industrial devices in the production system to obtain the processed real - time data.

[0096] In a specific implementation, the multi - protocol compatibility module in the edge layer is responsible for uniformly processing the real - time data of all industrial devices in the production system transmitted through different types of industrial communication protocols (such as EtherCAT, OPC UA, Profinet, etc.) to obtain the processed real - time data. Thus, even if the manufacturers, types of industrial devices are different, or the communication protocols used are different, the multi - protocol compatibility module can ensure the integration of data between them, thereby shielding the protocol differences between different industrial devices, enabling the predictive maintenance system to seamlessly integrate multi - brand and multi - type devices, and achieving cross - industrial device and cross - manufacturer interconnection and interoperability.

[0097] Specifically, after obtaining the real - time data, the multi - protocol compatibility module in the edge layer can first extract useful target data (exemplarily, extract sensor readings, timestamps, etc. from EtherCAT frames; extract device status, alarm information, etc. from OPC UA messages); then, according to predefined criteria, map the target data to a unified data structure to obtain the processed real - time data, such as JSON format, XML format, etc. Exemplarily, the temperature data extracted from the EtherCAT frame can be mapped to {"sensor":"temperature","value":75,"timestamp":"2025 - 02 - 05T14:40:00Z"}; the device status data extracted from the OPC UA message can be mapped to {"device":"pump","status":"running","timestamp":"2025 - 02 - 05T14:40:00Z"}.

[0098] It should be noted that the multi - protocol compatibility module in the edge layer can also shield some real - time data. By shielding unnecessary real - time data, not only can the transmission load be reduced, bandwidth and storage resources be saved, but also the response speed and overall performance of the predictive maintenance system can be improved. In one example, to achieve data shielding, only the occurrence period of events of interest (i.e., the target time period) can be focused on, while ignoring the real - time data collected at other times. For example, focus on the real - time data generated during the startup or shutdown process of the target industrial device, while relaxing the real - time data generated during the stable operation of the target industrial device. It should be noted that this application does not limit the specific data shielding rules.

[0099] It should also be noted that the multi - protocol compatibility module in the edge layer can also eliminate the noise data in the real - time data of all industrial devices in the production system, and filter out the invalid data in the real - time data of all industrial devices in the production system.

[0100] It should also be noted that the multi - protocol compatibility module in the edge layer can be located in the industrial gateway. For this, the present application makes no limitation.

[0101] S306: The data processing module in the edge layer transfers the processed real - time data of all industrial devices in the production system to the data processing module in the central layer.

[0102] It should be noted that at this time, the processed real - time data of all industrial devices in the production system has undergone unified protocol processing, and the noise data has been eliminated, and the invalid data has been filtered.

[0103] S307: The data processing module in the central layer inputs the processed real - time data of all industrial devices in the production system into the second decision - making generation model located in the central layer, and obtains the global maintenance decision corresponding to the processed real - time data, so that the data processing module in the central layer can perform predictive maintenance on all industrial devices in the production system according to the global maintenance decision.

[0104] The second decision - making generation model located in the central layer not only considers the processed real - time data of a single industrial device, but also pays attention to the collaborative work and mutual influence of all industrial devices in the entire production system, so as to obtain a more accurate and comprehensive global maintenance decision. See Figure 5 , which is a training flow chart of a second decision - making generation model provided by an embodiment of the present application. The second decision - making generation model is specifically trained through the following process:

[0105] B1: Obtain the historical operation data of all industrial devices in the production system.

[0106] B2: Obtain the second historical failures of all industrial devices in the production system when operating under the historical operation data.

[0107] B3: Based on the second historical failures, determine the historical global maintenance decisions of the production system for solving the second historical failures.

[0108] It can be understood that the above steps B1 - B3 are similar to steps A1 - A3, and will not be elaborated here.

[0109] B4: Train a machine - learning model according to the historical operation data of all industrial devices in the production system, the historical global maintenance decisions, and the influence characteristics between all industrial devices in the production system, and obtain the second decision - making generation model.

[0110] In addition to the historical operation data and historical global maintenance decisions of all industrial equipment in the production system, it is also necessary to consider the influence characteristics among all industrial equipment in the production system, such as the physical connections, data transmission relationships, and logical associations among all industrial equipment in the production system. Exemplarily, if the first industrial equipment is a motor and the second industrial equipment is a pump, and the motor drives the pump, then the speed and torque of the motor will affect the output flow rate and pressure of the pump (i.e., the influence characteristics). If there is a problem with the motor, such as unstable rotation speed or decreased power, this will immediately be reflected in the working efficiency of the pump. Therefore, based on the historical operation data, historical global maintenance decisions of all industrial equipment in the production system, and the influence characteristics among all industrial equipment in the production system, a more accurate second decision generation model can be trained. Thus, the machine learning model can not only learn the behavior patterns of individual industrial equipment but also learn the actual situation of multiple industrial equipment working together.

[0111] It can be understood that during the training process of the machine learning model, first, the machine learning model is trained using initial parameters to generate preliminary predicted global maintenance decisions; second, the error between the predicted global maintenance decisions and the historical global maintenance decisions is calculated, and this error is quantified using a loss function; subsequently, based on the error feedback, the model parameters of the machine learning model are adjusted using the gradient descent algorithm to gradually reduce the value of the loss function, and the above process is repeated until the error between the predicted global maintenance decisions of the machine learning model and the historical global maintenance decisions is minimized, thereby obtaining a second decision generation model with high accuracy.

[0112] After obtaining the global maintenance decisions provided by the second decision generation model, the central layer will perform predictive maintenance on the entire production system according to these global maintenance decisions. Exemplarily, if the processed real-time data indicates unstable rotation speed or decreased power of the motor, the trained second decision generation model can generate corresponding global maintenance decisions based on the processed real-time data and the influence characteristics among all industrial equipment in the production system, such as checking the motor bearings and replacing them, and at the same time suggesting monitoring the output flow rate and pressure of the pump.

[0113] It should be noted that since the industrial equipment in the production system will change over time, such as the introduction of new industrial equipment, the adjustment of the process flow, or the change of working conditions, etc., these changes may affect the effectiveness of the existing second decision generation model. Therefore, it is necessary to update the second decision generation model regularly to make it better adapt to the new situation, maintain the prediction accuracy, and reduce the problem of prediction inaccuracy caused by model aging.

[0114] Thus, the predictive maintenance system provided by the embodiments of the present application performs more complex global data analysis and long-term prediction through the central layer, and the design of jointly sharing the policy generation task between the edge layer and the central layer not only avoids the centralization of computing load, but also significantly improves the computing efficiency. Specifically, the edge layer is responsible for real-time data collection and local maintenance decision generation. Since these policy generation tasks usually require quick response, executing them near the data source (i.e., near the device layer) can reduce latency and improve the reaction speed. The central layer focuses on generating more complex global maintenance decisions. It can collect the processed real-time data of all industrial devices in the production system aggregated from the edge layer and use powerful computing resources for in-depth analysis to identify long-term trends or correlation patterns across devices.

[0115] Furthermore, if all policy generations are concentrated on the cloud or the central server, with the expansion of the industrial scenario scale, it may lead to computing resource overload and form a performance bottleneck. Such a centralized architecture is difficult to handle a large number of concurrent requests, especially in the case of high-frequency data updates. However, by allocating some computing tasks to the edge layer in the present application, the burden on the central layer is reduced, enabling each layer to work efficiently within its capabilities, thereby improving the overall response speed and stability of the predictive maintenance system.

[0116] In practical applications, the choice between global maintenance and local maintenance depends on the following factors: First, if the interdependence between industrial devices in the production system is complex and the failure of one industrial device may affect other industrial devices, global maintenance should be selected; on the contrary, if the industrial devices are highly independent, local maintenance can be chosen. Second, during the production peak period, downtime needs to be minimized as much as possible, and local maintenance is given priority; while during the production off-season or regular maintenance period, global maintenance can be arranged. Third, based on the available human, material resources and budget situation, decide whether to conduct comprehensive maintenance or local maintenance. The present application does not limit the factors for specifically choosing global maintenance or local maintenance.

[0117] See Figure 6 This figure is a schematic diagram of another predictive maintenance system provided by the embodiments of the present application. As Figure 6 can be seen, the predictive maintenance system is mainly divided into three layers: the device layer, the edge layer, and the central layer. Each layer is responsible for different tasks and jointly realizes the efficient monitoring and maintenance of industrial devices.

[0118] The device layer includes PLC-1 of production line 1 and PLC-2 of production line 2. Among them, both PLC-1 of production line 1 and PLC-2 of production line 2 include vibration sensors, temperature sensors, pressure sensors, proximity sensors, and frequency converters. Among them, the vibration sensor is used to monitor the vibration of industrial equipment for detecting mechanical failures. The temperature sensor is used to measure the operating temperature of industrial equipment to prevent overheating problems. The pressure sensor is used to monitor the pressure of fluids or gases. The proximity sensor is used to detect the presence or position change of an object. The frequency converter is used to adjust the speed and torque of the motor to optimize energy use. These sensors are connected to the industrial gateway in the edge layer through different communication protocols. Specifically, the vibration sensor, temperature sensor, and pressure sensor transmit data through the Modbus protocol, the proximity sensor transmits data through the Profibus protocol, and the frequency converter transmits data through the USS protocol. In order to uniformly transmit the data of these different protocols to the edge layer, the industrial gateway located in the edge layer plays a key role in protocol conversion. The industrial gateway converts the data of the Modbus, Profibus, and USS protocols into the Profinet or EtherCAT protocol. These two protocols are commonly used high-speed, real-time communication protocols in the field of industrial automation, capable of supporting large-scale data transmission and real-time communication between devices. Through this protocol conversion, the industrial gateway unifies the heterogeneous data of different sensors and industrial devices into the Profinet or EtherCAT protocol format, ensuring the efficient transmission and processing of real-time data in the edge layer. This conversion not only solves the multi-protocol compatibility problem but also improves the efficiency and reliability of data transmission, providing a unified data foundation for subsequent edge computing and decision-making generation.

[0119] The edge layer includes an edge server and an industrial gateway, mainly responsible for the preliminary processing and analysis of real-time data, as well as solving communication problems between different industrial devices. Among them, the industrial gateway includes a multi-protocol compatibility and standardization conversion module, which is used to uniformly convert the data from different sensors into a standard format to solve the communication problem between different industrial devices and the edge layer. The edge server includes a data processing and analysis module, a local storage module, and a decision-making maintenance module. The data processing and analysis module is used to receive the data from the industrial gateway and perform preliminary processing (including data cleaning, feature extraction, anomaly detection, etc.), reducing the amount of data transmitted to the central layer and improving efficiency. The local storage module is used to temporarily store the processed data in the local database for real-time analysis and historical data reference. The decision-making maintenance module is used to generate maintenance decisions using the first decision-making generation model and provide specific maintenance suggestions.

[0120] The central layer includes data center servers / cloud computing. Among them, the global model training and update management module receives the processed real-time data from the edge layer and uses machine learning algorithms (such as LSTM, Transformer, etc.) to train and optimize the first decision-making generation model and the second decision-making generation model to ensure its effectiveness in practical applications. The model update and system management module is used to update the trained first decision-making generation model to the edge layer.

[0121] In summary, the embodiment of the present application provides a predictive maintenance system, which includes a device layer, an edge layer, and a central layer. The present application uses the first decision-making generation model located in the edge layer to analyze the real-time data of the target industrial equipment, and can identify potential faults of the target industrial equipment in advance, determine and take local maintenance decisions to avoid unexpected downtime of the target industrial equipment. This not only improves the reliability of the target industrial equipment, but also reduces unplanned time delays and maintenance costs. Alternatively, the present application uses the second decision-making generation model located in the central layer to analyze the processed real-time data of all industrial equipment in the production system. Considering the collaborative work situation of all industrial equipment in the entire production system, deeper potential faults can be discovered, and then more comprehensive global maintenance decisions can be determined and taken to maximize the overall productivity of the production system and minimize downtime.

[0122] See Figure 7 , which is a flowchart of a predictive maintenance method provided by the embodiment of the present application. This predictive maintenance method is applied to a predictive maintenance system, which includes a device layer, an edge layer, and a central layer. Among them, the device layer is communicatively connected to the edge layer, and the edge layer is communicatively connected to the central layer. The method includes:

[0123] S701: The device layer obtains the real-time data of the target industrial equipment in the production system.

[0124] S702: The device layer inputs the real-time data of the target industrial equipment into the edge layer so that the edge layer generates local maintenance decisions for the target industrial equipment.

[0125] S703: The edge layer maintains the target industrial equipment according to the local maintenance decision.

[0126] Alternatively, after executing step S701, the following steps S704-S706 are executed:

[0127] S704: The device layer inputs the real-time data of all industrial equipment in the production system into the edge layer so that the edge layer processes the real-time data.

[0128] S705: The edge layer inputs the processed real-time data into the central layer so that the central layer generates global maintenance decisions for all industrial equipment in the production system.

[0129] S706: The central layer performs maintenance on all industrial devices in the production system according to the global maintenance decision.

[0130] In some specific implementation manners, the target industrial device is any one industrial device in the production system, or multiple industrial devices located on the same production line in the production system, or multiple industrial devices in the production system whose similarity of operating states is higher than a similarity threshold, where the operating state includes any one or more of temperature, pressure, vibration frequency, current, and voltage.

[0131] In some specific implementation manners, the edge layer includes a first decision generation model; inputting the real-time data of the target industrial device into the edge layer to enable the edge layer to generate a local maintenance decision for the target industrial device, including: inputting the real-time data of the target industrial device into the first decision generation model to enable the first decision generation model to generate a local maintenance decision for the target industrial device.

[0132] In some specific implementation manners, the first decision generation model is trained in the following manner: obtaining the historical operation data of the target industrial device and the first historical fault that the target industrial device has during the operation under the historical operation data; determining, based on the first historical fault, the historical local maintenance decision of the target industrial device for solving the first historical fault; training a machine learning model according to the historical operation data and the historical local maintenance decision to obtain the first decision generation model.

[0133] In some specific implementation manners, the first decision generation model includes a first sub-model and a second sub-model; inputting the real-time data of the target industrial device into the edge layer to enable the edge layer to generate a local maintenance decision for the target industrial device, including: inputting the real-time data of the target industrial device into the first sub-model to obtain prediction fault information corresponding to the real-time data of the target industrial device; inputting the prediction fault information into the second sub-model to obtain a local maintenance decision for the target industrial device corresponding to the prediction fault information.

[0134] In some specific implementation manners, after obtaining the prediction fault information corresponding to the real-time data of the target industrial device, the method further includes: when the prediction fault information indicates that the fault degree of the target industrial device is higher than a preset degree threshold, triggering an alarm indication.

[0135] In some specific implementation manners, the central layer includes a second decision generation model; inputting the processed real-time data into the central layer to enable the central layer to generate a global maintenance decision for all industrial devices in the production system, including: inputting the processed real-time data into the second decision generation model to enable the second decision generation model to generate a global maintenance decision for all industrial devices in the production system.

[0136] In some specific implementation manners, the second decision-making generation model is trained as follows: Obtain the historical operation data of all industrial devices in the production system, and the second historical failures that all industrial devices in the production system have during the operation under the historical operation data; Based on the second historical failures, determine the historical global maintenance decisions used by the production system to solve the second historical failures; Train a machine learning model according to the historical operation data, the historical global maintenance decisions, and the influence characteristics among all industrial devices in the production system to obtain the second decision-making generation model.

[0137] In some specific implementation manners, after obtaining the real-time data of the target industrial device in the production system, the method further includes: When the real-time data of the target industrial device does not meet the preset conditions, trigger an alarm indication, where the preset conditions include any one or more of the real-time temperature data of the industrial device being lower than the preset temperature threshold, the growth rate of the real-time temperature data of the industrial device being lower than the first preset growth threshold, the real-time pressure data of the industrial device being lower than the maximum safety threshold and higher than the minimum safety threshold, the growth rate of the real-time pressure data of the industrial device being lower than the second preset growth threshold, the decrease rate of the real-time pressure data of the industrial device being lower than the preset decrease threshold, the real-time humidity data of the industrial device being lower than the preset humidity threshold, the growth rate of the real-time humidity data of the industrial device being lower than the third preset growth threshold, and the real-time flow data of the industrial device being higher than the preset flow threshold.

[0138] In some specific implementation manners, the edge layer processes the real-time data, including: The edge layer performs unified format processing on the real-time data of all industrial devices in the production system.

[0139] In summary, the present application discloses a predictive maintenance method. The present application uses the first decision-making generation model located in the edge layer to analyze the real-time data of the target industrial device, and can identify potential failures of the target industrial device in advance, determine and take local maintenance decisions to avoid unexpected downtime of the target industrial device. This not only improves the reliability of the target industrial device, but also reduces unplanned time delays and maintenance costs. Or, the present application uses the second decision-making generation model located in the central layer to analyze the processed real-time data of all industrial devices in the production system. Since the collaborative working conditions of all industrial devices in the entire production system are considered, deeper potential failures can be discovered, and then more comprehensive global maintenance decisions can be determined and taken to maximize the overall productivity of the production system and minimize downtime.

[0140] The embodiment of the present application further provides an electronic device, which includes: a memory for storing a computer program or computer instructions; a processor for executing the computer program or computer instructions stored in the memory, so that the electronic device executes the steps performed by the device layer, the edge layer, or the central layer in the above predictive maintenance method.

[0141] The embodiments of the present application also provide a computer-readable storage medium, in which a computer program or instruction is stored. When the computer program or instruction is run, the steps executed by the device layer, edge layer, or central layer in the above predictive maintenance method are implemented.

[0142] The embodiments of the present application also provide a computer program product. When the computer program product runs on a server, the server is enabled to implement the steps executed by the device layer, edge layer, or central layer in the above predictive maintenance method.

[0143] It should be understood that in the embodiments of the present application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B may be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or a similar expression means any combination of these items, including any combination of single item (one) or plural items (ones). For example, at least one (one) of a, b, or c may mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c may be single or multiple.

[0144] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0145] The steps of the method or algorithm described in combination with the embodiments disclosed in this article can be directly implemented by hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0146] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A predictive maintenance method, applied to a predictive maintenance system, wherein the predictive maintenance system comprises a device layer, an edge layer and a central layer, wherein the device layer is communicatively connected to the edge layer, and the edge layer is communicatively connected to the central layer, wherein: The method comprises: The device layer obtains real-time data of target industrial equipment in the production system; The device layer inputs the real-time data of the target industrial equipment into the edge layer, so that the edge layer generates a local maintenance decision for the target industrial equipment; The edge layer performs maintenance on the target industrial equipment according to the local maintenance decision; or, The device layer inputs the real-time data of all industrial devices in the production system into the edge layer, so that the edge layer processes the real-time data; The edge layer inputs the processed real-time data into the central layer so that the central layer generates a global maintenance decision for all industrial equipment in the production system; The central layer maintains all industrial equipment in the production system according to the global maintenance decision.

2. The method according to claim 1, characterized in that The target industrial equipment is any one industrial equipment in the production system, or multiple industrial equipment located on the same production line in the production system, or multiple industrial equipment in the production system whose operating status similarity is higher than a similarity threshold, wherein the operating status includes any one or more of temperature, pressure, vibration frequency, current and voltage.

3. The method according to claim 1 or 2, characterized in that: The edge layer includes a first decision generation model; the real-time data of the target industrial equipment is input into the edge layer so that the edge layer generates a local maintenance decision for the target industrial equipment, including: The real-time data of the target industrial equipment is input into the first decision generation model, so that the first decision generation model generates a local maintenance decision for the target industrial equipment.

4. The method according to claim 3, characterized in that The first decision generation model is trained in the following manner: Acquire historical operation data of the target industrial equipment and a first historical fault of the target industrial equipment operating under the historical operation data; Based on the first historical fault, determining a historical local maintenance decision of the target industrial equipment for solving the first historical fault; A machine learning model is trained according to the historical operation data and the historical local maintenance decisions to obtain a first decision generation model.

5. The method according to claim 3 or 4, characterized in that: The first decision generation model includes a first sub-model and a second sub-model; the inputting the real-time data of the target industrial equipment into the edge layer so that the edge layer generates a local maintenance decision for the target industrial equipment includes: Inputting the real-time data of the target industrial equipment into the first sub-model to obtain predicted fault information corresponding to the real-time data of the target industrial equipment; The predicted fault information is input into the second sub-model to obtain a local maintenance decision corresponding to the predicted fault information for the target industrial equipment.

6. The method according to claim 5, characterized in that After performing the step of obtaining predicted fault information corresponding to the real-time data of the target industrial equipment, the method further includes: When the predicted fault information indicates that the fault degree of the target industrial equipment is higher than a preset threshold, an alarm indication is triggered.

7. The method according to claim 1, characterized in that The central layer includes a second decision generation model; the processed real-time data is input into the central layer so that the central layer generates a global maintenance decision for all industrial equipment in the production system, including: The processed real-time data is input into the second decision generation model so that the second decision generation model generates a global maintenance decision for all industrial equipment in the production system.

8. The method according to claim 7, characterized in that The second decision generation model is trained in the following manner: Acquire historical operation data of all industrial equipment in the production system, and second historical faults of all industrial equipment in the production system operating under the historical operation data; Based on the second historical fault, determining a historical global maintenance decision of the production system for solving the second historical fault; A machine learning model is trained based on the historical operating data, the historical global maintenance decisions, and the impact characteristics between all industrial equipment in the production system to obtain a second decision generation model.

9. The method according to claim 1, characterized in that: The edge layer processes the real-time data, including: The edge layer processes the real-time data of all industrial equipment in the production system in a unified format.

10. An electronic device, characterized in that: The electronic device comprises: Memory for storing computer programs or computer instructions; A processor, configured to execute a computer program or computer instructions stored in the memory, so that the electronic device executes the steps performed by the device layer, edge layer or central layer in the predictive maintenance method as described in any one of claims 1 to 9.

Images