Device operation state data self-adaptive processing method based on cloud-side-end collaborative architecture

Abstract

The invention discloses an equipment operation state data self-adaptive processing method based on a cloud-side-end collaborative architecture, and the method comprises the steps: firstly collecting various types of data from different equipment or parts, and evaluating the analysis demands of each type of data based on the collected data, including the needed calculation time, the time delay requirement and the calculation power size; the requirements of different data on real-time performance are understood; according to characteristics and analysis requirements of data, after hardware systems of a cloud end, an edge end and an equipment end are configured in advance and real-time state data of equipment are received, configuration of the cloud end, the edge end and the equipment end is directly called to conduct self-adaptive matching of data processing modes, and the self-adaptive matching comprises a local computing mode, a far-end computing mode and a data uplink / model downlink mode; then the computing tasks are distributed to the cloud end, the edge end and the equipment end; by calculating the execution result of the task, the feedback result is applied to the decision of the monitoring system, and the changing state and requirement are responded in time, so that the requirements of high performance, low time delay and accurate perception of the system are met.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent device monitoring, and particularly relates to an adaptive processing method for device operation state data based on a cloud-edge-end collaborative architecture. Background Art

[0002] Real-time monitoring systems are widely used in the industrial field, such as in many aspects of production processes, quality control, fault prediction and prevention, safety, maintenance management, energy efficiency, etc. To ensure that the system can effectively play its role, it needs to meet various requirements such as real-time performance, accuracy, and reliability. However, during the process of real-time monitoring of devices, a large amount of monitoring information such as device operation states, control states, and mining environments needs to be uploaded in real time, and the types of data are numerous. Remote visual intervention also requires a large amount of video data to be uploaded in real time. The real-time transmission of a series of data occupies a large amount of channel bandwidth, resulting in an increase in data transmission delay and a decrease in the reliability of the monitoring system.

[0003] With the rapid development of new-generation information technologies such as 5G, industrial Internet, and time-sensitive networks, as well as various intelligent sensors, the state monitoring of devices is continuously iterating and upgrading towards the intelligent stage. Cloud computing and edge computing, as important tools for big data analysis, have a profound impact on the establishment of real-time monitoring systems. Cloud computing focuses on centralized large-scale computing and storage resources, while edge computing focuses on real-time data processing and decision-making near the data source. By the collaborative cooperation of the two computing models of cloud computing and edge computing, a way that takes into account both high performance and low latency can be provided to integrate computing capabilities. The cloud-edge-end collaborative architecture, as Figure 1 shown, realizes more efficient data processing and application service provision by distributing computing resources and services to the network edge. The cloud, edge, and end cooperate with each other, and can provide faster, safer, and more reliable services. At the same time, it can also play an important role in reducing network transmission delay, reducing network load, and protecting user privacy. Combining the cloud-edge-end collaborative architecture, giving full play to the advantages of the cloud side, edge side, and device side, reasonably allocating data analysis tasks, and matching suitable computing capabilities to different tasks. Summary of the Invention

[0004] To solve the technical problems that the massive and diverse types of data generated during the operation of the device lead to an increase in latency during the transmission process, and some computing resources are not effectively utilized during the data analysis process, resulting in a decrease in computing efficiency, which directly affects the real-time performance, accuracy, and reliability of the real-time monitoring system, the present invention aims to provide a new data adaptive processing method based on a cloud-edge-device collaborative architecture. By identifying the data type and adaptively allocating data processing tasks according to the different computing powers required for data analysis, matching the corresponding data processing modes, and reasonably allocating the computing capabilities of the cloud, edge, and device sides, the advantages of the three are fully utilized to meet the requirements of "accurate" and "agile" perception of the monitoring system.

[0005] Starting from the hardware configurations of the cloud, edge, and device sides, after receiving the real-time status data of the device, the present invention analyzes the requirements of the data calculation tasks, selects the appropriate data processing mode, and invokes the pre-configured hardware of the cloud, edge, and device sides for adaptive selection of the mode and efficient processing of the data. To solve the technical problems, the technical solution of the present invention is as follows:

[0006] An adaptive processing method for device operation status data based on a cloud-edge-device collaborative architecture, the processing method comprising:

[0007] First, collect various types of data from different devices or components. Based on the collected data, intelligently evaluate the analysis requirements of each type of data, including the required calculation time, i.e., latency requirements, and the computing power, and understand the real-time requirements of different data. According to the characteristics and analysis requirements of the data, select the matching data processing mode, including the local calculation mode, remote calculation mode, data upload / model download mode. Then, allocate the calculation tasks to the cloud, edge, and device sides. Through the execution results of the calculation tasks, apply the feedback results to the monitoring system decision-making and respond in a timely manner to the changing status and requirements.

[0008] Further, for the local calculation mode, the resources participating in the local calculation mainly include the sensors and numerical calculation units configured on the device side. In this mode, the sensors transmit the collected real-time data to the numerical calculation unit to trigger the threshold comparison calculation task, and form a decision opinion based on the calculation results.

[0009] Further, in the local calculation mode, for the large amount of conventional data generated by the device during the operation state, considering the timeliness of data processing, directly process the data on the device side. By setting thresholds, directly judge the data on the device side and form a decision opinion.

[0010] Further, for the remote calculation mode, the resources participating in the remote calculation include: sensors, digital calculation units, and calculation gateways on the device side, edge computing servers on the edge side, cloud computing servers and historical databases on the cloud platform;

[0011] The real-time data collected by the sensor, after data preprocessing such as data denoising and outlier removal, is distributed by the gateway. Among them, it is distributed to the historical database for data backup, and according to the computing power requirements, it is distributed to the cloud computing server or the edge-side server to complete complex index calculation tasks, and decision-making opinions are formed based on the calculation results.

[0012] Furthermore, in the remote computing mode:

[0013] After the intelligent sensor collects data, it directly performs denoising preprocessing on the data locally, converts the data into low-noise data, and uploads the low-noise data to the edge-side computing unit through the communication protocol for index calculation and real-time data decision-making;

[0014] Or after the intelligent sensor converts the data into low-noise data, it uploads the low-noise data to the cloud database and the cloud platform computing unit respectively through the communication protocol. On the one hand, the low-noise data is backed up in the database of the cloud platform for offline feature extraction and offline analysis of the data. On the other hand, high-computing-power complex index calculations are performed on the low-noise real-time data in the computing unit of the cloud platform;

[0015] Or large-scale offline processing and complex calculations are performed on the multi-dimensional historical data stored in the cloud on the cloud platform computing unit.

[0016] Furthermore, in the data uplink / model downlink mode: This mode is divided into two stages: data uplink and model downlink. In the data uplink stage, the data reflecting the device status is uploaded to the historical database of the cloud platform, and the cloud computing server establishes a relationship fitting model between the status and data, and between data and data. In the model downlink stage, the trained model is downloaded to the edge-side server or the numerical calculation unit of the device end, and calculation results are obtained based on the real-time collected data or the key data set, and decision-making opinions are formed.

[0017] Furthermore, in the data uplink / model downlink mode:

[0018] After the intelligent sensor collects data, it performs basic index calculation locally, extracts feature data, uploads the feature data to the historical database of the cloud platform through the gateway for data fusion, extracts the key data set, and then uses the multi-dimensional historical data stored in the historical database of the cloud to establish a data relationship model in the cloud platform computing unit. Then the model is downloaded to the edge-side computing unit, cooperates with the key data set obtained by data fusion and performs data processing, and uses the fitting relationship between the data to verify the key data set, so as to further monitor the device operation status;

[0019] Alternatively, after preprocessing the original data at the device side, the relationship model fitted by the cloud from the corresponding historical data can be directly downloaded to the intelligent sensor, and the model can be directly used to screen and monitor abnormal data locally.

[0020] The hardware configuration methods for the cloud, edge, and device sides based on the above data processing mode are as follows:

[0021] Starting from the analysis requirements of device status data, different data can adopt local analysis at the device side, cloud analysis, edge-device collaboration, cloud-device collaboration, and cloud-edge-device collaboration respectively.

[0022] Device-side hardware configuration: Micro embedded computing unit, communication module.

[0023] Edge-side hardware configuration: Edge computing server, medium and small storage devices, communication module.

[0024] Cloud-side hardware configuration: High-performance cloud computing server, cloud database, cloud data processing framework, high-bandwidth network interface.

[0025] When performing local analysis at the device side, the data is uploaded to the micro embedded computing unit for simple calculations such as threshold judgment. When performing cloud analysis, the historical data is imported into the high-performance computing unit, and the cloud data processing framework is called for complex index calculations. When performing edge-device collaborative computing, the data is first uploaded to the micro embedded computing unit at the device side for data preprocessing, and then the preprocessed data is uploaded to the edge side for index calculations. When performing cloud-device analysis, the data is first uploaded to the micro embedded computing unit at the device side for data preprocessing, and then the processed data is uploaded to the high-performance computing platform in the cloud for complex index calculations. When performing cloud-edge-device collaborative computing, the data is first preprocessed in the micro embedded computing unit at the device side. The cloud uses the device historical status data and high-performance computing environment to build a model. Secondly, the model built in the cloud is downloaded to the edge computing unit or the micro embedded computing unit at the device side, and the preprocessed data uploaded is analyzed and calculated at the edge side and the device side respectively.

[0026] In actual use, according to the data type and analysis requirements, the above several computing units on the cloud, edge, and device sides can be adaptively called for data mining and analysis. According to different data types and analysis requirements, the system will intelligently select the appropriate computing mode. Specifically, when quick response to data is required, the device side or the edge side will be preferentially selected for computing; while for complex model building and large-scale data analysis, the cloud side will be selected for processing. By real-time evaluating the task requirements, the system can adaptively call the computing resources of the device side, edge side, or cloud side to ensure that the computing tasks are efficiently executed in the most suitable hardware environment.

[0027] Compared with the prior art, the advantages of the present invention are as follows:

[0028] The present invention proposes an adaptive processing method for device operation status data based on a cloud-edge-device collaborative architecture. According to different data types generated by different devices and components, as well as the time and computing power required for different data analysis and calculations, it adaptively matches the data processing mode, reasonably distributes the computing tasks to the cloud, edge, and device sides, ensures the full utilization of computing resources, thereby improving the efficiency of data processing and transmission, and meeting the requirements of the online monitoring system for real-time, accuracy, reliability, and other aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is the cloud-edge-device collaborative architecture;

[0030] Figure 2 is the specific working process of the present invention;

[0031] Figure 3 is the first method for processing the status data of a roadheader under the cloud-edge-device collaborative architecture;

[0032] Figure 4 is the second method for processing the status data of a roadheader under the cloud-edge-device collaborative architecture;

[0033] Figure 5 is the third method for processing the status data of a roadheader under the cloud-edge-device collaborative architecture;

[0034] Figure 6 is the fourth method for processing the status data of a roadheader under the cloud-edge-device collaborative architecture;

[0035] Figure 7 is the fifth method for processing the status data of a roadheader under the cloud-edge-device collaborative architecture;

[0036] Figure 8 is the sixth method for processing the status data of a roadheader under the cloud-edge-device collaborative architecture;

[0037] Figure 9 are six data flow patterns under the cloud-edge-device collaborative architecture. DETAILED DESCRIPTION OF THE INVENTION

[0038] The following describes the specific embodiments of the present invention in conjunction with the embodiments:

[0039] It should be noted that the structures, ratios, sizes, etc. shown in this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the implementation conditions of the present invention. Any modification of the structure, change of the proportional relationship or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0040] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of clear narration, and are not used to limit the scope of implementation of the present invention. The change or adjustment of their relative relationship, without substantial change in the technical content, should also be regarded as the scope of implementation of the present invention.

[0041] Embodiment 1:

[0042] This embodiment proposes a method for adaptively processing device operation state data. According to different data types generated by different devices and components, as well as the time and computing power required for different data analysis and calculations, it adaptively matches the data processing mode, reasonably distributes the computing tasks to the cloud, edge, and device sides, ensures the full utilization of computing resources, thereby improving the efficiency of data processing and transmission, and meeting the requirements of the online monitoring system in terms of real-time performance, accuracy, reliability, etc.

[0043] For this method, first, based on the cloud-edge-device collaborative architecture and the computing tasks, the approximate time required to match the corresponding numerical computing capabilities, that is, based on the latency requirements and computing power requirements, three modes for the computing power allocation problem are proposed: local computing mode, remote computing mode, data uplink / model downlink. These three modes include the following six specific data processing methods.

[0044] (1) Local computing mode: The resources participating in local computing mainly include sensors and numerical computing units configured on the device side. The sensors in this mode transmit the collected real-time data to the numerical computing unit to trigger the threshold comparison computing task, and form a decision opinion based on the calculation result, specifically as Method 1.

[0045] (2) Remote computing mode: The resources participating in remote computing mainly include sensors, digital computing units, and computing gateways on the device side, edge computing servers on the edge side, cloud computing servers and historical databases on the cloud platform. The real-time data collected by the sensors undergoes data preprocessing such as data denoising and outlier removal, and is distributed by the gateway. Among them, it is distributed to the historical database for data backup, and is distributed to the cloud computing server or the edge-side server according to the computing power requirements to complete complex index calculation tasks, and form a decision opinion based on the calculation result, specifically as Method 2, Method 3, and Method 6

[0046] (3) Data Upload / Model Download Mode: This mode is divided into two stages: data upload and model download. In the data upload stage, the data reflecting the device status is uploaded to the historical database of the cloud platform, and the cloud computing server establishes a relationship fitting model between the status and data, and between data and data. In the model download stage, the trained model is downloaded to the edge-side server or the numerical calculation unit of the device end, and the calculation results are obtained based on the real-time collected data or the key data set to form a decision opinion, specifically as in Method 4 and Method 5.

[0047] Six specific data processing methods:

[0048] Method 1: For some relatively unimportant data generated when the device is in operation, although the impact on the device operation is weak, in order to ensure the safety of the staff, its monitoring is still indispensable. In this mode, since the data change rate is slow, the amount of data generated is relatively small, and the data characteristics are relatively single, not much processing is required. Therefore, it can be directly processed at the device end, and by setting thresholds and other methods, the data can be directly judged and a decision can be formed at the device end.

[0049] Method 2: After the intelligent sensor collects the data, it directly performs noise reduction, preprocessing, etc. on the data locally, converts the data into low-noise data, and uploads the data to the edge-side computing unit through the communication protocol for some index calculations and real-time data decisions.

[0050] Method 3: Different from Mode 2, in this mode, after the intelligent sensor converts the data into low-noise data, it uploads the low-noise data to the cloud database and the cloud platform computing unit respectively through the communication protocol. On the one hand, the data is backed up in the database of the cloud platform for offline feature extraction and offline analysis, and on the other hand, some complex indexes that require high computing power are calculated for the low-noise real-time data in the computing unit of the cloud platform.

[0051] Method 4: After the intelligent sensor collects the data, it performs basic index calculations locally, extracts the feature data, uploads the feature data to the edge side through the gateway for data fusion, and extracts the key data set. Then, using the historical data stored in the historical database of the cloud, a data relationship model is established in the cloud platform computing unit, and then the model is downloaded to the edge-side computing unit, combined with the key data set obtained by data fusion for data processing, and the fitting relationship between the data is used to verify the key data set to further monitor the device operation status.

[0052] Method 5: After preprocessing the original data at the device end, the relationship model fitted by the corresponding historical data in the cloud is directly downloaded to the intelligent sensor, and the model is directly used to screen and monitor abnormal data locally.

[0053] Method 6: Use the multi-dimensional historical data stored in the cloud to perform large-scale offline processing and some complex calculations on the computing units of the cloud platform.

[0054] The working principle of the device operation data adaptive processing method proposed in this paper is as follows:

[0055] First, the status data of the device is collected by various types of sensors. After the data is obtained, it is distributed by the gateway. According to the processing requirements of the obtained data (including the requirements of the monitoring system for data analysis latency and the computing power requirements for different data processing) and the computing power distribution of the cloud-edge-device collaborative architecture, the data processing tasks are distributed to the computing devices with matching computing power. Then, the computing units of the cloud, edge, and device sides are respectively called to process and analyze the corresponding data, so as to maximize the use of the computing resources under the collaborative architecture, quickly complete the computing tasks of each part, and quickly feedback the computing results to the system, so as to make decisions in a timely manner. The specific work process is as Figure 3 shown.

[0056] The above six methods can generally describe the distribution of the computing capabilities of the intelligent sensor among the cloud, edge, and device sides. By the requirements of data processing for time and various indicators, and the computing power required for model establishment, the computing tasks are reasonably allocated, so as to maximize the use of the computing capabilities of the three to improve the efficiency of data processing and the accuracy, reliability, and real-time performance of the monitoring system.

[0057] Example 2:

[0058] A device operation status data adaptive processing method based on the cloud-edge-device collaborative architecture mainly has two functions: adaptive matching of data processing modes and organization and allocation of computing power.

[0059] As a large-scale electromechanical device in coal mines, the roadheader is mainly used to complete the tunneling of underground roadways. Taking the operation status monitoring of the roadheader as an example, the specific application of this sensor is demonstrated. In this invention, various data of the cutting part of the roadheader are mainly collected, including hydraulic oil temperature, motor vibration, speed, temperature, and component load data, etc. The sensor adaptively judges the data according to different requirements of various data, selects a suitable data processing line, and distributes the data to the cloud and edge sides respectively through the gateway, allocates reasonable data processing tasks, and fully invokes the data analysis resources to improve the data processing efficiency.

[0060] As Figure 4As shown, for data such as hydraulic oil temperature, when analyzing this type of data, the computing power required for this type of data is relatively small. Therefore, under the cloud-edge-device collaborative architecture, it is chosen to directly perform data judgment and processing at the device end. The temperature data collected by the intelligent temperature sensor is directly judged locally through the controller according to the temperature threshold set for the sensor, and then a decision is formed.

[0061] As Figure 5 shown, for the motor vibration signal, since it usually contains multiple frequency components, when analyzing the vibration signal, the interference of various noises causes the vibration signal to be subject to certain interference and distortion, which has a certain impact on the signal acquisition, transmission, and processing, and may even lead to misjudgment. Therefore, during the processing, it is first necessary to perform noise reduction processing on the vibration signal to remove noise interference, improve the signal-to-noise ratio, and improve the signal characteristics. Since the computing power required for noise reduction processing is relatively small, the signal noise reduction processing can be directly performed locally at the device end. Then, the obtained low-noise signal is uploaded to the edge side through a communication protocol for calculating relevant indicators, including time-domain indicators, frequency-domain indicators, time-frequency domain indicators, statistical feature indicators, etc. Compared with the device end, the computing power and efficiency of the edge side are greatly improved, and it can effectively extract features. The edge side can also provide high resolution, capture the instantaneous changes and subtle frequency components of the vibration signal, and has strong real-time performance and self-adaptability. Under the condition of meeting accuracy, the computing efficiency can be guaranteed. By calculating the above four indicators, the operating state of the motor can be further judged based on the changes of the indicators compared with the normal situation, and then a decision can be made.

[0062] As Figure 6 shown, this mode is still based on the analysis of the vibration signal generated by the motor. Different from the previous step, this mode uses the more powerful computing power, computing resources, and storage capacity of cloud computing to analyze and mine the low-noise vibration signal.

[0063] Calculate the model recognition indicators in the cloud, such as autoregressive model, autoregressive moving average model (ARMA). Use the uploaded low-noise data to establish an autoregressive moving average model in the cloud. Considering the autocorrelation and moving average characteristics of the vibration signal comprehensively, the dynamic characteristics of the vibration signal can be described more accurately. It can also predict the vibration signal at future moments, thus helping to prevent equipment failures or perform maintenance.

[0064] As Figure 7As shown, taking the motor speed data as an example, some simple index calculations of the speed are carried out at the device end, and then the above-mentioned indexes are uploaded to the edge side for data fusion to obtain a key index data set. Utilizing the powerful computing power of the cloud, the historical data set is called, the vibration data and the speed data are fused, and a vibration-speed relationship model is established. An algorithm is used to fit the relationship model of vibration and speed, and then the model is downloaded to the edge side. At the edge side, the relationship fitting of the index data is carried out to screen out outliers. At the same time, the real-time motor speed data and fault data are uploaded to the cloud to update the vibration-speed relationship model to ensure the accuracy of the judgment result.

[0065] As Figure 8 shown, taking the motor temperature as an example, the motor temperature is different from the oil temperature. As an important engineering parameter, the motor temperature directly affects the performance, life and reliability of the equipment. And its requirement for real-time performance is extremely high. It is necessary to quickly send the decision to the execution unit to avoid unnecessary losses caused by the damage of the motor. In this mode, all operations are executed at the device end to ensure the real-time performance requirements. First, it is still data collection and preprocessing. For the motor temperature, some lightweight calculation tasks can be carried out at the device end. At the same time, the motor temperature is closely related to the motor speed and vibration. Using the historical data set, a temperature-speed-vibration model is established in the cloud, the relationship among the three is fitted, and the model is directly sent to the device end to judge the real-time temperature, and feedback and decision are made when abnormal temperature is detected. Then, according to the fitting relationship among the three, the vibration and speed data under abnormal temperature are further analyzed to carry out mutual diagnosis and verification among the three.

[0066] As Figure 9 shown, when monitoring the cutting part of a roadheader, it is often necessary to analyze a series of indexes such as equivalent stress, strain, deformation amount, stiffness, etc. of the cutting part under the condition of load loading to verify the design reliability and safety of the periphery of the cutting part. Due to reasons such as the complexity of the structure, the finite element analysis method is usually used to integrate the control data, node data, element data, boundary condition data, etc. of the components to carry out structural analysis on the cutting part. Finite element analysis usually takes a lot of time, and there are many nodes and elements in the model; at the same time, different types of loads such as static load, dynamic load, and thermal load all require different computing resources. Throughout the process of finite element analysis, a large amount of computing resources are usually required. In order to complete the analysis faster, it may be necessary to use a dedicated high-performance computing (HPC) system or cloud-based computing resources. Therefore, for such operations, it is taken as a separate computing mode, and the historical data stored in the cloud platform is called to perform high-performance complex calculations. For example, the above-mentioned method of using finite element analysis to carry out structural analysis of the cutting part, and the reliability and safety of the components are judged through offline analysis. At the same time, the size of the structure can be optimized to save material costs.

[0067] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0069] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0070] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0071] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments. Within the knowledge of those of ordinary skill in the art, various changes can be made without departing from the spirit of the present invention.

[0072] Many other changes and modifications can be made without departing from the concept and scope of the present invention. It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for adaptively processing device operation status data based on a cloud-edge-end collaborative architecture, characterized in that: The processing method comprises: First, various types of data are collected from different devices or components. Based on the collected data, the analysis requirements of each type of data are intelligently evaluated, including the required computing time (i.e. latency requirements) and computing power, and the real-time requirements of different data are understood. According to the characteristics of the data and the analysis requirements, a matching data processing mode is selected, including local computing mode, remote computing mode, and data uplink / model downlink mode. The computing tasks are then assigned to the cloud, edge, and device ends. Through the execution results of the computing tasks, the feedback results are applied to the monitoring system decisions, and the changing status and needs are responded to in a timely manner.

2. According to claim 1, a method for adaptively processing device operation status data based on a cloud-edge-end collaborative architecture is characterized in that: The local computing mode, in which resources involved in local computing mainly include sensors and numerical computing units configured on the device side, the sensors in this mode transmit the collected real-time data to the numerical computing unit to trigger the threshold comparison computing task, and form a decision opinion based on the computing results.

3. According to claim 2, a method for adaptively processing device operation status data based on a cloud-edge-end collaborative architecture is characterized in that: In local computing mode, a large amount of routine data generated by the device when it is in operation is processed directly on the device side, taking into account the timeliness of data processing. By setting thresholds, the data is directly judged and decisions are made on the device side.

4. According to claim 1, a method for adaptively processing device operation status data based on a cloud-edge-end collaborative architecture is characterized in that: In the remote computing mode, the resources involved in the remote computing include: sensors, digital computing units and computing gateways on the device side, edge computing servers on the edge side, and cloud computing servers and historical databases on the cloud platform; The real-time data collected by the sensor is pre-processed by data noise reduction and outlier removal, and then distributed by the gateway. Among them, it is distributed to the historical database for data backup, and distributed to the cloud computing server or edge server according to the computing power requirements to complete the complex indicator calculation tasks, and the decision opinions are formed according to the calculation results.

5. According to claim 4, a method for adaptively processing device operation status data based on a cloud-edge-end collaborative architecture is characterized in that: In remote computing mode: After the smart sensor collects the data, it directly performs noise reduction preprocessing on the data locally, converts the data into low-noise data, and uploads the low-noise data to the edge computing unit through the communication protocol for indicator calculation and real-time data decision-making; Or after the smart sensor converts the data into low-noise data, it uploads the low-noise data to the cloud database and the cloud platform computing unit through the communication protocol. On the one hand, the low-noise data is backed up in the cloud platform database for offline feature extraction and offline analysis of the data. On the other hand, the computing unit of the cloud platform calculates complex indicators with high computing power for the low-noise real-time data. Or use the multi-dimensional historical data stored in the cloud to perform large-scale offline processing and complex calculations on the cloud platform computing unit.

6. The method for adaptively processing device operation status data based on cloud-edge-end collaborative architecture according to claim 1, characterized in that: The data uplink / model downlink mode: This mode is divided into two stages: data uplink and model downlink. In the data uplink stage, data reflecting the device status is uploaded to the historical database of the cloud platform, and the cloud computing server establishes a fitting model for the relationship between status and data, and between data and data. In the model downlink stage, the trained model is decentralized to the numerical calculation unit of the edge server or device, and the calculation results are obtained based on the real-time collected data or key data sets to form decision opinions.

7. The method for adaptively processing device operation status data based on cloud-edge-end collaborative architecture according to claim 6, characterized in that: In the data uplink / model downlink mode: After collecting data, the smart sensor calculates basic indicators locally, extracts feature data, and uploads the feature data to the cloud platform's historical database through the gateway for data fusion, extracts key data sets, and then uses the historical data stored in the cloud's historical database to establish a data relationship model in the cloud platform computing unit. The model is then downloaded to the edge computing unit, and the key data sets obtained by data fusion are processed, and the fitting relationship between the data is used to verify the key data sets, thereby further monitoring the equipment operation status; Or after preprocessing the raw data on the device side, the relational model fitted by the corresponding historical data in the cloud is directly downloaded to the smart sensor, and the model is directly used to screen and monitor abnormal data locally.

Images