A remote control and fault diagnosis method and system for Internet of Things devices

Abstract

The present invention discloses a remote control and fault diagnosis method and system for an Internet of Things device, belonging to the field of remote control. The method comprises: arranging sensors around the device to collect environmental data, device log data of the device, and user behavior data, and transmitting the collected data to the device's Internet of Things framework edge gateway for processing; sending the calculated unified feature vector to the cloud, and at the same time building a real-time monitoring architecture in the edge gateway based on the unified feature vector; after the real-time monitoring architecture captures a health jump, the edge gateway issues a local warning signal according to the shutdown warning rule and uploads the data to the cloud; the cloud analyzes the fault and generates a repair strategy through a policy generation model; the cloud distributes the repair strategy generated by the policy generation model. The present invention analyzes and attributes the fault based on the feature vector and the historical health sequence, thereby improving the data preprocessing capability.

Description

Technical Field

[0001] The present invention relates to the field of remote control, and in particular to a remote control and fault diagnosis method and system for Internet of Things devices. Background Art

[0002] With the continuous development of industrial automation and information technology, the Internet of Things (IoT) has been widely adopted in the industrial sector. Traditional industrial equipment monitoring and management methods rely primarily on regular manual inspections and offline sampling and analysis. These methods lack real-time performance, comprehensiveness, and cost-effectiveness, making them difficult to meet the requirements of modern industry for environmental quality monitoring. While IoT technology can apply various data collection and control sensors to all aspects of industrial production, processing and fusing heterogeneous multi-source data from multiple sensors in complex industrial environments remains a significant challenge.

[0003] Existing technologies have the following major problems when using multi-source heterogeneous data for equipment status detection: First, multiple parts of the system are upgraded or replaced at different time scales, network connectivity may change, there are multiple noise sources, process operating parameters are highly variable, and there are nonlinear responses to abnormal operations, etc., which may cause sensing requirements to vary over time; second, previous industrial systems are not flexible in quickly and in real time configuring the sensed parameters and managing system variables such as intermittent network availability, and do not integrate data from offset systems into sensor planning and execution in real time; third, existing data processing methods often only target a single data source and cannot effectively integrate multi-source heterogeneous data, making it difficult to fully reflect the operating status of the equipment.

[0004] Therefore, there is an urgent need for an equipment status detection method that can collect and fuse multi-source heterogeneous data in real time to improve the real-time, accuracy and comprehensiveness of detection and meet the needs of modern industrial environment monitoring. Summary of the Invention

[0005] One of the purposes of the present invention is to provide a remote control and fault diagnosis method for Internet of Things devices to solve the problem in the existing technology that it is impossible to effectively integrate multi-source heterogeneous data and it is difficult to fully reflect the operating status of the device.

[0006] The present invention is implemented through the following technical solution, which provides a remote control and fault diagnosis method for an Internet of Things device, including the following steps: S1, arranging sensors around the device to collect environmental data around the device, collecting device log data and user behavior data of the device, transmitting the collected data to the device's Internet of Things framework edge gateway, and preprocessing and tensorizing the collected data; S2, sending the calculated unified feature vector to the cloud, and at the same time constructing a real-time monitoring architecture in the edge gateway based on the unified feature vector; S3, after the real-time monitoring architecture captures a health jump caused by user operation or environmental change, the edge gateway issues a local warning signal according to the shutdown warning rule, and uploads the unified feature vector at the triggering moment and the historical health sequence to the cloud; S4, the cloud analyzes and attributes the fault based on the received unified feature vector and the historical health sequence, and generates a repair strategy based on the constructed strategy generation model; S5, the cloud summarizes the fault based on the analysis results to construct a fault database, and issues the repair strategy generated by the strategy generation model, notifying the operator to prepare the resources and tools required to perform the repair operation, and after the maintenance is completed, uploading the repair operation summary report to the cloud for storage in the fault database.

[0007] Furthermore, S1 also includes the following sub-steps: S1.1, the collected environmental data, device log data and user behavior data are matrixed, where the device log matrix L, L∈R T×D , R is the real number domain, T is the time point, D is the log entry; the environmental data matrix E, E∈R T×S×P , S is the sensor type, P is the spatial position of the sensor; the user behavior data is integrated into a discrete user behavior sequence U, U={u1, u2, ..., u k}, each event u k contains the operation type and timestamp, and k is the number of events. S1.2. Use the Tucker decomposition method to decompose the device log matrix, environmental data matrix, and user behavior sequence, extract potential features, and maintain the consistency of the time dimension to construct a unified feature space. Then, align the time dimension and force the sharing of the time factor matrix to ensure that all data is synchronized on the time axis, as shown in the following formula:

[0008] ,

[0009] Among them, U time is the time factor matrix, It means finding the parameters that make the objective function take the minimum value, L is the device log matrix, G log is the core tensor of the device log matrix, U is the user behavior sequence, and U feature is the characteristic factor matrix, is the sensor type factor matrix, is the spatial position factor matrix, is the core tensor of the environment data matrix, × n is the n-module product of tensors; S1.3 projects the core tensors of each data source into a unified space to construct a joint feature space, as shown in the following formula: , where F is the unified eigenvector; α, β, γ are weight coefficients; The linear transformation function of the core tensor of the device log matrix; is the linear transformation function of the core tensor of the environment data matrix; is the linear transformation function of the transition probability matrix.

[0010] Furthermore, the weight coefficient is adaptively calculated through the attention mechanism.

[0011] Furthermore, the device log matrix L is decomposed into: , where G log is the core tensor of the device log matrix; U time is the time factor matrix, used to capture the time pattern; U feature is the characteristic factor matrix used to map the log dimension to a low-dimensional space.

[0012] Furthermore, the environmental data matrix E is decomposed into: ,in, is the core tensor of the environment data matrix; U time is the time factor matrix, used to capture the time pattern; is the sensor type factor matrix; is the spatial position factor matrix. In the above two formulas, × n is the n-modular product of tensors.

[0013] Furthermore, the user behavior sequence U, a discrete event sequence, is converted into a transition probability matrix M u :

[0014] , where M U ∈R A×A , A is the number of operation types.

[0015] Furthermore, building a real-time monitoring architecture includes: building a health function H(t) for each device by combining user operations and environmental factors, performing real-time health calculations using the health function H(t), capturing health jumps caused by user operations or environmental changes, monitoring and analyzing the data, and providing local warnings.

[0016] Furthermore, the real-time monitoring architecture uses a stochastic differential equation to describe the continuous evolution of device health. The stochastic differential equation is shown below: ,in, is the drift function, is the health of the device, which is between [0, 1]. When H(t) = 0, it means that the device is completely faulty, and when H(t) = 1, it means that the device is in the best state. is the differential symbol, which represents a small change or increment of a variable. Device health A small change in time t, is a small increment of time t, is the random variation of the random external shock at time t; is the unified eigenvector function, For user operation input, The diffusion term represents the effect of random diffusion, is the random diffusion coefficient, is a random external shock.

[0017] Furthermore, the drift function is shown as follows: ,in, is the natural attenuation term, is the natural aging coefficient, is the initial equipment health; is the environmental impact item, is the environmental sensitivity enhancement coefficient, tanh is the hyperbolic tangent function, W F is the feature weight vector, corresponding to the weight of the unified feature vector function, For user intervention items, Gain factor for user operation.

[0018] Furthermore, the diffusion term is expressed as follows: ,in, is the basic noise intensity constant, is the health of the equipment, and n is the nonlinear coefficient.

[0019] Furthermore, S4 includes the following sub-steps: S4.1. Calculating the health degradation gradient based on the historical health sequence. When the degradation rate exceeds a threshold, it is determined to be rapid degradation, triggering diagnostic analysis. The health degradation gradient is calculated by the following formula:

[0020] ,in, is the Laplace operator, is the gradient of health decline, For device health, is the time interval, is the health of the device after the time interval; S4.2, the physical quantity of the unified eigenvector is used as the device parameter node, the core tensor of the environment data matrix in step S1 is used as the environment node, and the high-frequency operation in the user behavior transfer matrix is ​​used as the operation node. The industrial knowledge is embedded to construct a causal graph and the candidate root dependent variables are screened; S4.3, for each candidate root dependent variable X i , combined with the parameters of the stochastic differential equation, we can calculate the expected change in health after the intervention and measure the candidate root dependent variable X by the following formula: i Impact on system health: ,in, For X i After the intervention, the expected change in system health is: H represents the system health, E is the expected value, is the external intervention operator, Represents variable X i Normal value that should be in; Represents X i The current actual value.

[0021] Furthermore, the policy generation model is constructed by mixed integer linear programming, including the objective function and constraints: define the binary variable y j and the continuous variable x k , where y j ∈{0, 1}, y j =1, it means to perform intervention or adjustment on the corresponding candidate root dependent variable. j =0 means no intervention or adjustment operation is performed on the corresponding candidate root dependent variable; k Used for parameter adjustment; the objective function is constructed as shown below: , where c j is the operating cost, y j is a binary variable, d k is the parameter adjustment cost, x k is a continuous variable, is the expected downtime cost; j is the index of the operation; the constraints include causal constraints:

[0022] when , which means the dependent variable must be intervened, is the set of root dependent variables; physical safety constraints: , i is the feature index, is the system behavior generated under the feature index, For safety boundaries, is the weight coefficient; health recovery constraint: ,in, is the safety margin, θ is the safety redundancy, Repair time for the equipment.

[0023] On the other hand, the present invention provides a remote control and fault diagnosis system for Internet of Things devices, which includes: a processor; a memory storing a computer program. When the computer program is executed by the processor, the remote control and fault diagnosis method for Internet of Things devices as described above is implemented.

[0024] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0025] 1. The present invention improves the overall computing speed and realizes rapid fault diagnosis by preprocessing and tensorizing the collected multi-source heterogeneous data in the edge gateway, thereby solving the problem that the existing technology lacks real-time performance and flexibility. At the same time, tensor decomposition is performed through the Tucker decomposition method to extract the potential features of each data source while retaining the consistency of the time dimension, so that the core tensor dimension can be greatly compressed to reduce the subsequent computing load, improve the computing speed and facilitate rapid judgment, and effectively solve the problem that the existing technology cannot effectively integrate multi-source heterogeneous data.

[0026] 2. The present invention uses an attention mechanism to adjust the weight of the data source according to the current device status. For example, when the temperature is abnormal, the weight of the environmental data is increased, thereby improving the ability to fully reflect the operating status of the device and overcoming the shortcomings of the existing technology that it is difficult to fully reflect the operating status of the device. It also describes the continuous evolution of the device health through stochastic differential equations, captures the health jumps caused by user operations or environmental changes, monitors and analyzes the data, and provides local warnings.

[0027] 3. The present invention analyzes and attributes faults based on feature vectors and historical health sequences, generates repair strategies, and builds a fault database to facilitate subsequent detection and maintenance, thereby improving data preprocessing and cross-validation capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0029] Figure 1 This is a flow chart of the method provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0031] Example 1

[0032] This embodiment discloses a remote control and fault diagnosis method for IoT devices. This method combines IoT technology with data collection to achieve intelligent and automated status monitoring and abnormal fault diagnosis of factory equipment, as well as analysis and repair of abnormalities.

[0033] Figure 1 A flow chart of the remote control and fault diagnosis method in this embodiment is shown. It can be seen from the figure that this embodiment includes the following steps:

[0034] Step 1: Temperature sensors, vibration sensors, noise sensors, humidity sensors, and air pressure sensors (multiple sensors) are deployed around devices within the same IoT framework to collect environmental data around the devices.

[0035] Collect device logs from devices within the same IoT framework. In this embodiment, the device logs should include: device operating status parameters (for example, voltage (V), current (A), speed (RPM), vibration amplitude (mm / s), etc.), device fault codes, operation records, and other data related to device operation.

[0036] At the same time, relevant data involving user behavior operations such as device parameter adjustment, forced restart / emergency stop, maintenance mark, firmware OTA upgrade, backup / restore configuration file, network configuration change, etc. are collected.

[0037] The collected data is transmitted to the edge gateway of the IoT framework, where it is preprocessed and tensorized.

[0038] Specifically, in this embodiment, a multi-source data fusion algorithm is configured in the edge gateway to perform refined processing on the collected multi-data. The edge gateway pre-processes the data to improve the overall computing speed and achieve rapid fault diagnosis.

[0039] The specific steps include the following:

[0040] 1) First, data is divided into device log data, environmental data collected by sensors, and user behavior data according to different data types, and these three types of data are processed separately.

[0041] Among them, the device log matrix L, L∈R T×D , R is the real number field, T is the time point, D is the log entry, and T×D is used to represent D types of log entries within T time points.

[0042] Environmental data matrix E, E∈R T×S×P, T is the time point, S is the sensor type, and P is the spatial position of the sensor.

[0043] L∈R T×D The device log matrix is ​​a real matrix with time points T as rows and log entries D as columns. T×S×P The environmental data matrix is ​​a real matrix of size T × S × P. The elements of these matrices are all real numbers, so R is used to indicate that their elements come from the real number domain.

[0044] Integrate user behavior data into a discrete user behavior sequence U, U={u1,u2, ..., u k}, each event u k It contains the operation type (such as restart, configuration modification) and timestamp. It should be noted that the definition of U is U={u1,u2, ...,u k}, in this definition, k represents the number of events, meaning that U contains k events, and u k It refers specifically to the kth event.

[0045] 2) Tucker decomposition is then used to decompose the device log matrix L, the environmental data matrix E, and the user behavior sequence U, extracting the potential features of each data source while preserving the consistency of the time dimension and constructing a unified feature space.

[0046] Decompose the device log matrix L into:

[0047] ,

[0048] Among them, G log is the core tensor of the device log matrix; U time is the time factor matrix, used to capture the time pattern; U feature is the characteristic factor matrix used to map the log dimension to a low-dimensional space.

[0049] Decompose the environmental data matrix E into:

[0050] ,

[0051] in, is the core tensor of the environment data matrix; U time is the time factor matrix, used to capture the time pattern; is the sensor type factor matrix; is the spatial position factor matrix. In the above two formulas, × n is the n-modular product of tensors.

[0052] Convert the user behavior sequence U, a discrete event sequence, into a transition probability matrix M u :

[0053]

[0054] Among them, M U ∈R A×A , A is the number of operation types.

[0055] Then, the time dimension is aligned to force the sharing of the time factor matrix to ensure that all data are synchronized on the time axis, as shown in the following formula:

[0056]

[0057] ,

[0058] 3) Then project the core tensors of each data source into a unified space to construct a joint feature space as shown below:

[0059] ,

[0060] Among them, F is the unified eigenvector; α, β, γ are weight coefficients; The linear transformation function of the core tensor of the device log matrix; is the linear transformation function of the core tensor of the environment data matrix; is the linear transformation function of the transition probability matrix.

[0061] In this embodiment, the weight coefficient can be obtained by adaptive calculation through the attention mechanism. For example, the weight coefficient α can be calculated by the following formula:

[0062] ,

[0063] in, is the attention parameter matrix.

[0064] It's important to note that in the above steps, the Tucker decomposition method is used to decompose the tensor, extracting the potential features of each data source while maintaining consistency in the time dimension. Furthermore, an attention mechanism is used to adjust the weight of the data source based on the current device state (for example, increasing the weight of environmental data when the temperature is abnormal). This allows the core tensor dimension to be compressed to 1 / 10 the size of the original data, reducing the subsequent computational load and increasing the speed of operation for faster judgment.

[0065] Step 2: Send the calculated unified feature vector F to the cloud, and build a real-time monitoring architecture based on the unified feature vector F in the edge gateway.

[0066] Specifically, building a real-time monitoring architecture includes, after obtaining a unified feature vector F through data collection, building a health function H(t) for each device by combining user operations and environmental factors, performing real-time health calculations using the health function H(t), capturing health jumps caused by user operations or environmental changes, monitoring and analyzing the data, and providing local warnings.

[0067] In this embodiment, a stochastic differential equation is used to describe the continuous evolution of the device health. Specifically, the stochastic differential equation can be expressed as follows:

[0068] ,

[0069] in, is the drift function, is the health of the device, which is in the interval [0, 1], where H(t) = 0 indicates a complete failure of the device, and H(t) = 1 indicates that the device is in the best condition. It is the unified feature vector output by the stage-, which contains various features of the current environment of the device. User operation input, used to indicate the user's direct intervention behavior on the health of the device. The diffusion term represents the effect of random diffusion, is the random diffusion coefficient. is a random external shock with independent increments and normal distribution properties. This term captures the impact of random environmental noise on device health. is the differential symbol, which represents a small change or increment of a variable. Device health A small change in time t, is a small increment of time t, is the random variation of the random external shock at time t;

[0070] In the stochastic differential equation disclosed in this embodiment, the drift function is composed of three parts: natural attenuation of the device, environmental characteristics, and user operation. The specific form of the drift function of this embodiment can be shown as follows:

[0071] ,

[0072] in, It is a natural attenuation term used to describe the natural aging of equipment over time. is the natural aging coefficient, It is an environmental impact item, which is used to reflect the impact of the external environment on the health of the device, and uses a weight vector to adjust the impact of different environmental characteristics. is the environmental sensitivity enhancement coefficient, which is used to characterize the impact of environmental feature changes on device health. tanh is the hyperbolic tangent function, which is used to provide a smooth nonlinear mapping, making the impact of input features on health more continuous and controllable. F is the feature weight vector, corresponding to the weight of the unified feature vector function. Adjusting the weight can change the impact of different features on health. F(t) is a unified feature vector containing various characteristics of the device's current environment, such as ambient temperature, current, vibration, etc. , is the user intervention item, Gain coefficient for user operation, User operation input indicates the user's direct intervention in the health of the device.

[0073] It's important to note that the drift function in this example dynamically describes the deterministic rate of change of device health by combining three primary factors: natural attenuation, environmental impact, and user intervention. These parameters work together to accurately simulate the evolution of device health over time under different conditions, enabling real-time prediction of device health and maintenance planning.

[0074] In the stochastic differential equation disclosed in this embodiment, the diffusion term can be expressed as follows:

[0075] ,

[0076] in, is the basic noise intensity constant, which is a fixed value used to determine the baseline noise level. is the device health. k is a nonlinear coefficient that controls how the diffusion term changes with H(t). For example, when k = 2, the diffusion term decreases with the square of the health; when k = 1, the diffusion term decreases linearly with the health.

[0077] It should be noted that the diffusion term in this embodiment describes the impact of random fluctuations or uncertainty in the system on health. By introducing a random factor (typically Brownian motion or a Wiener process), changes in health are affected not only by the deterministic drift term but also by random interference. This random interference can originate from unforeseen factors such as external environmental fluctuations and internal noise. Introducing the diffusion term helps simulate the actual behavior of devices in different health states. For example, when device health is low, it is more sensitive to external environmental changes or internal noise, and the model can more accurately reflect this characteristic. This allows the entire real-time monitoring architecture to more realistically simulate the evolution of device health H(t), especially when health is low, where this volatility is more pronounced. This not only helps more accurately predict device failure risks but also provides a more reliable basis for developing equipment maintenance and preventive measures.

[0078] Step 3: After the real-time monitoring architecture detects a health jump caused by user operations or environmental changes and issues an alert, the edge gateway issues a local alert signal based on the downtime alert rules and uploads F(t) at the time the alert was triggered, along with the historical health sequence, to the cloud.

[0079] In this embodiment, the shutdown warning rules may be as shown in Table 1, a local warning rule table. The edge gateway may provide a local warning according to the warning level in Table 1.

[0080] Table 1. Local warning rules table

[0081]

[0082] Step 4: The cloud analyzes and attributes the fault based on the received F(t) and the historical health sequence, and generates a repair strategy based on the constructed strategy generation model.

[0083] Specifically, in this embodiment, analyzing and attributing the fault using F(t) and the historical health sequence includes the following sub-steps:

[0084] 1) First, the health degradation gradient is calculated based on the historical health sequence. When the decline rate exceeds the threshold, it is determined to be rapid degradation, triggering diagnostic analysis.

[0085] Specifically, in this embodiment, the health degree decline gradient can be calculated by the following formula:

[0086] ,

[0087] in, is the Laplace operator, is the gradient of health decline, For device health, is the time interval, The device health after the time interval has passed.

[0088] 2) The physical quantity of the unified eigenvector is used as the device parameter node, the core tensor of the environmental data matrix in step 1 is used as the environmental node, and the high-frequency operations in the user behavior transfer matrix are used as the operation node. The industrial knowledge is embedded in the construction to obtain a causal graph and screen out the candidate root dependent variables.

[0089] Specifically, we can integrate typical industrial fault libraries into the resulting causal graph and quickly screen candidate root dependent variables based on maintenance work order knowledge.

[0090] 3) For each candidate root dependent variable X i , combined with the parameters of the stochastic differential equation, back-calculate the expected change in health after the intervention.

[0091] Specifically, the candidate root dependent variable X can be measured by the following formula i Impact on system health.

[0092] ,

[0093] in, For X i The expected change in system health after intervention (setting it to normal value). If this value is large, it means that X i It has a significant impact on system health and may be the root cause of the problem. H represents system health and E is the expected value (i.e., average value).

[0094] It should be noted that in the above formula, do is an external intervention operator, which means external intervention on the system. Forced to be set This is similar to simply observing Natural The do operator forces changes to the system to see causal effects. Represents X i The normal value that should be in, that is, theoretically, the device or system is in the best state, X i The value that should be. Represents X i The current actual value.

[0095] To calculate The steps taken are as follows:

[0096] First freeze X i Set it to the normal value instead of the current possible abnormal value. Simulate its state under normal circumstances without being affected by the current environment.

[0097] Then the stochastic differential equation model is used to simulate the evolution of the system. i Recalculate the system's health after intervention H'. Compare and observe the difference between the current health H and the health H' obtained after the simulated intervention. Take H' as the expected value of health after intervention. .

[0098] Finally, calculate all candidate variables X i Changes in health caused by ,choose The variable with the largest absolute value means that this variable has the greatest impact on changes in health and is likely to be the root cause of the problem.

[0099] It should be noted that the above steps can quickly identify the root variables that have the greatest impact on system health. This helps determine the optimal repair strategy to effectively address anomalies in the system and provides an accurate diagnostic method for complex system fault diagnosis.

[0100] In this embodiment, the strategy generation model is constructed by mixed integer linear programming, which specifically includes:

[0101] 1) First, define the binary variable y j and the continuous variable x k , where y j ∈{0, 1},y j =1, it means to perform intervention or adjustment on the corresponding candidate root dependent variable. j =0 means no intervention or adjustment operation is performed on the corresponding candidate root dependent variable; k Used for parameter adjustment.

[0102] 2) Then the objective function is constructed as shown below:

[0103] ,

[0104] Among them, c j is the operating cost, y j is a binary variable, d k is the parameter adjustment cost, x k is a continuous variable, is the expected downtime cost (failure time predicted by the stochastic differential equation model).

[0105] 3) The constraints of the objective function include:

[0106] Causality constraints:

[0107] when , which means the root dependent variable must be intervened.

[0108] Physical security constraints:

[0109] , represents the system behavior generated under each feature index i in order to ensure the safe operation of the system Do not exceed the safety boundary .

[0110] Health recovery constraints:

[0111] ,

[0112] Health recovery constraints are used to ensure the health of the system after repair At least the safety level θ plus the safety margin δ should be met.

[0113] It should be noted that the policy generation model in this embodiment comprehensively determines the optimal repair solution by considering operational costs, parameter adjustment costs, and downtime, ensuring the comprehensiveness and efficiency of the repair strategy. The model incorporates causal constraints, physical safety constraints, and health recovery constraints to ensure that repair operations do not pose a safety hazard to the long-term operation of the system. Through constraint and objective function optimization, this model can minimize device downtime, improve device availability and operational efficiency, and flexibly adapt to IoT device failures of varying types and complexities, demonstrating its high versatility.

[0114] Step 5: The cloud summarizes the faults based on the analysis results and builds a fault database to facilitate subsequent detection and maintenance. The cloud also sends the repair strategy generated by the strategy generation model to the corresponding operators, notifying them to prepare the resources and tools required to perform the repair operation.

[0115] Operators execute repair operations in an orderly manner according to the generated repair strategy, and keep detailed records of each repair step, including the time, performer, and specific operation content. The repair operation summary report is then uploaded to the cloud and stored in the fault database. The fault is extracted and added to the typical industrial fault database or formed into industrial knowledge to facilitate subsequent model judgment.

[0116] Example 2

[0117] This embodiment discloses a remote control and fault diagnosis system for Internet of Things devices.

[0118] The remote control and fault diagnosis system for the Internet of Things device in this embodiment includes: a processor and a memory, in which a computer program is stored. When the computer program is executed by the processor, the remote control and fault diagnosis method for the Internet of Things device in Example 1 can be implemented.

[0119] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A remote control and fault diagnosis method for an Internet of Things device, characterized in that: The remote control and fault diagnosis method includes: S1. Place sensors around the device to collect environmental data around the device, device log data, and user behavior data. The collected data is transmitted to the edge gateway of the IoT framework of the device, and the collected data is preprocessed and tensorized; S2. Send the calculated unified feature vector to the cloud, and build a real-time monitoring architecture based on the unified feature vector in the edge gateway; S3. When the real-time monitoring architecture detects a health jump caused by user operations or environmental changes, the edge gateway issues a local warning signal based on the downtime warning rules and uploads the unified feature vector at the time of the warning trigger and the historical health sequence to the cloud. S4. The cloud analyzes and attributes the fault based on the received unified feature vector and historical health sequence, and generates a repair strategy based on the constructed strategy generation model. S5. The cloud summarizes the faults based on the analysis results and builds a fault database. It then distributes the repair strategy generated by the strategy generation model, notifying the operator to prepare the resources and tools required to perform the repair operation. After the inspection is completed, the repair operation summary report is uploaded to the cloud and stored in the fault database. Said S1 comprises: S1.

1. Matrix the collected environmental data, device log data, and user behavior data. Among them, the device log matrix L, L∈R T×D , R is the real number domain, T is the time point, and D is the log entry; Environmental data matrix E, E∈R T×S×P , S is the sensor type, P is the spatial position of the sensor; Integrate user behavior data into discrete user behavior sequences U, U={u1, u2, ..., u k }, each event u k contains the operation type and timestamp, and k is the number of events; S1.

2. Use the Tucker decomposition method to decompose the device log matrix, environmental data matrix, and user behavior sequence, extract potential features, and retain the consistency of the time dimension to construct a unified feature space. Then, the time dimension is aligned to force the sharing of the time factor matrix to ensure that all data are synchronized on the time axis, as shown in the following formula: ， Among them, U time is the time factor matrix, It means finding the parameters that make the objective function take the minimum value, L is the device log matrix, G log is the core tensor of the device log matrix, U is the user behavior sequence, and U feature is the characteristic factor matrix, is the sensor type factor matrix, is the spatial position factor matrix, is the core tensor of the environment data matrix, × n is the n-module product of tensors; S1.3 projects the core tensors of each data source into a unified space to construct a joint feature space. As shown in the following formula: ， Among them, F is the unified eigenvector; α, β, γ are weight coefficients; The linear transformation function of the core tensor of the device log matrix; is the linear transformation function of the core tensor of the environment data matrix; is the linear transformation function of the transfer probability matrix; The construction of the real-time monitoring architecture includes: Combine user operations and environmental factors to build the health function H(t) of each device, and perform real-time health calculations through the health function H(t). Capture health changes caused by user operations or environmental changes, monitor and analyze data, and provide local warnings; The real-time monitoring architecture uses a stochastic differential equation to describe the continuous evolution of device health. The stochastic differential equation is shown below: ， in, is the drift function, is the health of the device, which is between [0, 1]. When H(t) = 0, it means that the device is completely faulty, and when H(t) = 1, it means that the device is in the best state. is the differential symbol, which represents a small change or increment of a variable. Device health A small change in time t, is a small increment of time t, is the random variation of the random external shock at time t; is the unified eigenvector function, For user operation input, The diffusion term represents the effect of random diffusion, is the random diffusion coefficient, is a random external shock; The S4 includes: S4.

1. Calculate the health degradation gradient based on the historical health sequence. If the degradation rate exceeds a threshold, it is determined to be rapid degradation, triggering diagnostic analysis. The health decline gradient is calculated by the following formula: ， in, is the Laplace operator, is the gradient of health decline, For device health, is the time interval, is the health of the device after the time interval; S4.2: Use the physical quantities of the unified eigenvector as device parameter nodes, the core tensor of the environmental data matrix in S1.1 as environment nodes, and the high-frequency operations in the user behavior sequence as operation nodes. Embed the industrial knowledge to construct a causal graph and screen candidate root dependent variables. S4.

3. For each candidate root dependent variable X i , combined with the parameters of the stochastic differential equation, to calculate the expected change in health after the intervention. The candidate root dependent variable X is measured by the following formula i Impact on system health: ， in, For X i After the intervention, the expected change in system health is: H represents the system health, E is the expected value, is the external intervention operator, Represents X i Normal value that should be in; Represents X i The current actual value.

2. The remote control and fault diagnosis method of the Internet of Things device according to claim 1, characterized in that: The weight coefficient is obtained by adaptive calculation through the attention mechanism.

3. The remote control and fault diagnosis method of the Internet of Things device according to claim 1, characterized in that: The drift function is shown below: ， in, is the natural attenuation term, is the natural aging coefficient, is the initial equipment health; is the environmental impact item, is the environmental sensitivity enhancement coefficient, tanh is the hyperbolic tangent function, W F is the feature weight vector, corresponding to the weight of the unified feature vector function, For user intervention items, Gain factor for user operation.

4. The remote control and fault diagnosis method of the Internet of Things device according to claim 1, characterized in that: The diffusion term is expressed as follows: ， in, is the basic noise intensity constant, is the health of the equipment, and n is the nonlinear coefficient.

5. The remote control and fault diagnosis method of the Internet of Things device according to claim 1, characterized in that: The strategy generation model is constructed by mixed integer linear programming, including the objective function and constraints: Define the binary variable y j and the continuous variable x k , where y j ∈{0, 1}, y j =1, it means to perform intervention or adjustment on the corresponding candidate root dependent variable. j =0 means no intervention or adjustment operation is performed on the corresponding candidate root dependent variable; k Used for parameter adjustment; The objective function is constructed as follows: ， Among them, c j is the operating cost, y j is a binary variable, d k is the parameter adjustment cost, x k is a continuous variable, is the estimated downtime cost, j is the index of the operation; Constraints include: Causality constraints: when , which means the dependent variable must be intervened, is the set of root dependent variables; Physical security constraints: , i is the feature index, is the system behavior generated under the feature index, For safety boundaries, is the weight coefficient; Health recovery constraints: ， in, is the safety margin, θ is the safety redundancy, Equipment repair time, For time, For health.

6. A remote control and fault diagnosis system for Internet of Things devices, characterized in that: The remote control and fault diagnosis system includes: processor; The memory stores a computer program, and when the computer program is executed by the processor, the remote control and fault diagnosis method of the Internet of Things device according to any one of claims 1 to 5 is implemented.

Images