Equipment fault detection method based on Internet of Things and data driving
By combining the Internet of Things and deep learning, the fusion and collaborative analysis of multi-sensor data has been achieved, solving the problem of information silos in traditional equipment fault detection and improving the accuracy of fault detection and the operating efficiency of the production line.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2026-01-15
- Publication Date
- 2026-04-17
AI Technical Summary
Traditional equipment fault detection methods rely on isolated sensor data processing, resulting in information silos, a lack of multi-information fusion, weak anti-interference capabilities, and a high probability of false alarms and missed alarms, making it difficult to achieve accurate identification and collaborative control at the system level.
By integrating multiple sensors using IoT technology, and through the collaborative efforts of environmental monitoring, equipment status monitoring, fault detection, and control modules, combined with time-frequency analysis and deep learning models for data preprocessing and detection, cross-sensor data fusion and collaborative analysis are achieved.
It improves the accuracy and reliability of equipment fault detection, reduces the false alarm rate, and enables multi-dimensional and accurate fault location and type identification of equipment status, thereby improving the production efficiency of the production line and the level of equipment health management.
Smart Images
Figure CN121879299A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of fault detection, and specifically relates to a device fault detection method based on the Internet of Things and data-driven approaches. Background Technology
[0002] Traditional equipment fault detection typically relies on single, isolated detection methods, such as diagnosing bearing faults solely through vibration analysis or determining motor overload solely through temperature monitoring. In this model, each sensor operates independently, forming "information silos" with a lack of effective correlation and coordination between data. The workload for information processing is enormous and inefficient. Technicians often need to manually sift through massive amounts of discrete data to identify abnormal signals, which not only consumes a lot of manpower and time but also easily leads to key fault characteristics being overlooked due to information overload.
[0003] Of particular concern is that this isolated analysis approach severely neglects the potential organic connections between various detection information. For example, abnormal vibration may, together with temperature rise and pressure fluctuation, constitute a comprehensive characterization of a certain fault mode. However, traditional methods treat these three factors separately, resulting in the idle and wasted resources of a large amount of cross-domain and multi-source information, failing to fully explore their deep correlation value through information fusion. This "seeing the trees but not the forest" limitation means that fault diagnosis often remains at the level of local symptoms, making it difficult to construct an overall state profile at the system level, and even more so, failing to meet the needs of accurate identification and coordinated control of the operational status of complex equipment systems.
[0004] Furthermore, isolated data acquisition and processing methods have weak anti-interference capabilities. Single sensor signals are highly susceptible to environmental noise, electromagnetic interference, or their own drift. Without cross-validation and complementary correction from multi-source information, the probability of false alarms and missed alarms increases significantly, further reducing the accuracy and reliability of fault detection. In today's increasingly integrated and intelligent industrial systems, this fragmented detection mode has become a bottleneck in improving equipment health management, urgently requiring a transformation to an intelligent operation and maintenance paradigm that integrates multi-information and systematic diagnosis. Summary of the Invention
[0005] To address the problems existing in the background technology, this invention provides an equipment fault detection method based on the Internet of Things and data-driven approaches. This method solves the technical problems of isolated data acquisition and processing methods with weak anti-interference capabilities, lack of multi-information fusion and systematic diagnosis in existing technologies. Through the synergistic effect of sensors, monitoring modules, fault detection modules and control modules, efficient production line equipment fault detection is achieved.
[0006] The technical solution adopted in this invention is: I. A device fault detection method based on the Internet of Things and data-driven approach: S1. In the target monitoring area, deploy and install the production line environment monitoring module, equipment status monitoring module, fault detection module, control module, several environmental sensors and several equipment status sensors.
[0007] S2. The production line environment monitoring module obtains environmental data around the equipment from the installed environmental sensors; the equipment status monitoring module obtains equipment status data from the installed equipment status sensors.
[0008] S3, the production line environment monitoring module and the equipment status monitoring module preprocess the acquired data and then transmit them together to the fault detection module.
[0009] S4. The fault detection module performs fault detection on the received data, obtains the detection results, and transmits the detection results to the control module.
[0010] S5. The control module adjusts the equipment on the production line in real time based on the test results.
[0011] The environmental sensors include an environmental humidity sensor, an environmental temperature sensor, and an environmental image recognition sensor; the equipment status sensors include an equipment vibration sensor, an equipment pipeline pressure sensor, an equipment temperature sensor, an equipment image recognition sensor, and a contact displacement sensor.
[0012] The preprocessing involves sequentially performing A / D conversion, data calibration, and filtering on the data. The A / D conversion includes four steps performed sequentially: sampling, holding, quantization, and encoding.
[0013] The formula for the filtering algorithm used in the filtering process is set as follows: in, yes The state at any given moment, yes The state at any given moment, It is based on the previous state The prediction results obtained at any time The previous state The optimal result at any given time. Here is the state transition matrix. yes The transpose of the matrix, It is a control input matrix. It is the control input vector. yes The corresponding covariance, yes The corresponding covariance, It is the process noise covariance matrix. For the current moment The optimal estimate; For gain, yes The corresponding covariance, It is the observation matrix. yes The transpose of the matrix, For a single model and single measurement, the identity matrix is used. , It is the observation noise covariance matrix. yes The actual observation vector at time t.
[0014] The fault detection module includes a time-frequency processing submodule and a model detection submodule. The time-frequency processing submodule receives data transmitted from the production line environment monitoring module and the equipment status monitoring module, processes the data using a time-frequency analysis algorithm to obtain time-frequency graph data, and transmits the time-frequency graph data to the model detection submodule. The model detection submodule then performs detection on the time-frequency graph data to obtain the detection result.
[0015] The model detection submodule uses a pre-trained convolutional neural network.
[0016] II. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.
[0017] 3. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above method.
[0018] The beneficial effects of this invention are: 1. The method of this invention uses Internet of Things (IoT) technology to perform unified data processing and detection on production line environmental data and equipment status data collected by multiple sensors, realize cross-sensor data fusion and collaborative analysis, obtain more accurate, stable and reliable conclusions, reduce losses caused by equipment failures, and improve the production efficiency of the production line.
[0019] 2. This invention uses time-frequency analysis technology to preprocess sensor data, obtaining multi-dimensional and comprehensive data information. It can clearly see the changes in the corresponding frequency components at the moment of the fault, thereby accurately locating the fault location and type and improving the accuracy of equipment fault detection. Preliminary processing of sensor data can reduce the inaccuracy of data caused by inaccurate measurement and improve the accuracy of the description of system status.
[0020] 3. This invention employs a ResNet-50 residual neural network combined with a SAM spatial attention module to extract features and perform classification prediction on time-frequency maps. The multiple convolutional layers of the ResNet-50 residual neural network enable multi-scale feature extraction, providing excellent data features for fault detection; the SAM spatial attention module suppresses noise bands in the time-frequency map, increases the weight of effective signal regions, improves the accuracy of equipment fault detection, and reduces the learning of redundant background. Attached Figure Description
[0021] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0022] The present invention will now be described in more detail with reference to the accompanying drawings and embodiments. However, the present invention is not limited thereto. For those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention. Contents not described in detail in this specification are prior art known to those skilled in the art.
[0023] like Figure 1 As shown, the equipment fault detection method of this embodiment includes the following steps: S1. In the target monitoring area, deploy and install the production line environment monitoring module, equipment status monitoring module, fault detection module, control module, several environmental sensors and several equipment status sensors.
[0024] Environmental sensors include environmental humidity sensors, environmental temperature sensors, and environmental image recognition sensors; equipment status sensors include equipment vibration sensors, equipment pipeline pressure sensors, equipment temperature sensors, equipment image recognition sensors, and contact displacement sensors.
[0025] An environmental humidity sensor collects environmental humidity data of the target monitoring area, an environmental temperature sensor collects environmental temperature data of the target monitoring area, and an environmental image recognition sensor collects multi-angle environmental image sensing data of the target monitoring area.
[0026] Equipment vibration sensors collect vibration data from the equipment; equipment pipeline pressure sensors collect pressure data from the equipment pipelines; equipment temperature sensors collect temperature data from the equipment; equipment image recognition sensors collect image sensing data from the equipment; and contact displacement sensors collect displacement data from the equipment.
[0027] All data collected by all the sensors at the same time are aggregated to obtain the equipment fault detection data at a single moment.
[0028] S2. The production line environment monitoring module obtains environmental data around the equipment from the installed environmental sensors; the equipment status monitoring module obtains equipment status data from the installed equipment status sensors.
[0029] S3, the production line environment monitoring module and the equipment status monitoring module preprocess the acquired data and then transmit them together to the fault detection module.
[0030] Preprocessing involves sequentially performing A / D conversion, data calibration, and filtering on the data. A / D conversion includes four steps performed sequentially: sampling, holding, quantization, and encoding.
[0031] The production line environmental monitoring module performs A / D conversion and data calibration on the environmental data sequentially, while the equipment status monitoring module performs A / D conversion and data calibration on the equipment status data sequentially. Both modules convert the continuously changing analog signals collected by sensors into discrete digital signals to facilitate subsequent data processing. Finally, the data is calibrated to ensure accuracy and reliability.
[0032] The filtering process employs a filtering algorithm to estimate the true state of the system using a series of observational data, under conditions of uncertainty and noise. The formula for the filtering algorithm is as follows: in, yes The state at any given moment, yes The state at any given moment, It is based on the previous state The prediction results obtained at any time The previous state The optimal result at any given time. Here is the state transition matrix. yes The transpose of the matrix, It is a control input matrix. It is the control input vector. yes The corresponding covariance, yes The corresponding covariance, It is the process noise covariance matrix. For the current moment The optimal estimate; For gain, yes The corresponding covariance, It is the observation matrix. yes The transpose of the matrix, For a single model and single measurement, the identity matrix is used. , It is the observation noise covariance matrix. yes The actual observation vector at time t.
[0033] S4. The fault detection module performs fault detection on the received data, obtains the detection results, and transmits the detection results to the control module.
[0034] The fault detection module includes a time-frequency processing submodule and a model detection submodule. The time-frequency processing submodule receives data transmitted from the production line environment monitoring module and the equipment status monitoring module, processes the data using a time-frequency analysis algorithm to obtain time-frequency graph data, and transmits the time-frequency graph data to the model detection submodule. The model detection submodule then performs detection on the time-frequency graph data to obtain the detection results.
[0035] Using time-frequency analysis algorithms, the signal is decomposed into frequency components and their corresponding amplitudes at different time points, obtaining the signal's time and frequency information. This makes the data not only contain information about how the signal changes over time, but also the distribution of the signal at different frequencies, making it a multi-dimensional data that facilitates subsequent data classification and prediction.
[0036] The model detection submodule uses a pre-trained convolutional neural network.
[0037] In practice, the model that combines the ResNet-50 residual neural network with the SAM spatial attention module is used when pre-training the convolutional neural network.
[0038] The ResNet-50 residual neural network consists of five stages. The first stage (Stage 1) is a cascaded 7×7 convolutional and pooling layer, without residual connections. The following four stages (Stages 2-5) are composed of residual blocks with bottleneck structures, containing 3, 4, 6, and 3 bottleneck blocks respectively, for a total of 16. The structure of each bottleneck block is: 1x1 convolution (dimensionality reduction) -> 3x3 convolution -> 1x1 convolution (dimensionality increase), and spatial downsampling is performed in the first block of each stage in Stages 2-5.
[0039] This application does not impose any restrictions on the number of SAM spatial attention modules or the specific location of each SAM spatial attention module in the ResNet-50 residual neural network. The spatial attention mechanism can be applied to the two-dimensional plane of the time-frequency plot, and can simultaneously learn the joint distribution of key frequency bands and key time periods, enabling the model to learn which frequencies are important in which time periods, thereby increasing their importance and enhancing the classification and prediction capabilities of the convolutional network.
[0040] In this embodiment, four SAM spatial attention modules are used, with one SAM spatial attention module inserted in series after each stage in Stages 2-5. A hierarchical strategy is employed: the SAM convolutional kernel size after Stage 2 is set to 3*3, using a smaller receptive field to focus on fine details and small defects; the SAM convolutional kernel size after Stages 3 and 4 is set to 5*5, used to extract the shape features of the parts; and the SAM convolutional kernel size after Stage 5 is set to 7*7, used to extract the overall fault mode. The activation function is ReLU, the loss function is cross-entropy loss, the learning rate is set to 0.001, and training is performed using a learning rate scheduler. The batch size is set to 128.
[0041] During the pre-training process of convolutional neural networks: When collecting equipment fault detection data at various times, the system simultaneously labels whether a fault has occurred, including the specific equipment and its location.
[0042] The sliding window method is used to construct samples. The equipment fault detection data of N consecutive time moments form a sample window, which is used as the input of a single sample data in the network. The fault label sequence of the N consecutive time moments corresponding to the equipment fault detection data constitutes the label data of the single sample data.
[0043] Using a single sample data as input data, the convolutional neural network is pre-trained based on the input data and label data, thus completing the pre-training of the convolutional neural network.
[0044] S5. The control module adjusts the equipment on the production line in real time based on the test results.
[0045] After receiving the detection results, the fault detection module will prompt the operator to check the corresponding equipment and send the decision results to the corresponding control modules for adjustment to ensure the normal operation of the production line.
[0046] Furthermore, the fault detection module also sends data information to the upper-level server or platform for further processing and analysis, facilitating later backtracking.
[0047] The production line environmental monitoring module is responsible for collecting environmental data in the target monitoring area. The environmental data collected by the i-th environmental sensor is Di = {d1, d2, …, dk, …, dM}, where dk is the environmental data collected by the k-th environmental sensor.
[0048] The equipment status monitoring module is responsible for collecting equipment status data in the target monitoring area. The equipment status data collected by the i-th equipment status sensor is Pi = {p1, p2, …, pk, …, pM}, where pk is the equipment status data collected by the k-th equipment status sensor.
[0049] This embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.
[0050] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.
[0051] In summary, the method of the present invention effectively solves the problems in the prior art and provides an innovative solution for fault detection of production line equipment, which has significant practicality and economic benefits.
[0052] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.
Claims
1. A method for detecting equipment failure based on Internet of Things and data driving, characterized in that, include: S1. In the target monitoring area, deploy and install the production line environment monitoring module, equipment status monitoring module, fault detection module, control module, several environmental sensors and several equipment status sensors. S2. The production line environment monitoring module acquires environmental data around the equipment from the installed environmental sensors; the equipment status monitoring module acquires equipment status data from the installed equipment status sensors. S3, the production line environment monitoring module and the equipment status monitoring module preprocess the acquired data and then transmit them together to the fault detection module; S4. The fault detection module performs fault detection on the received data to obtain the detection result, and transmits the detection result to the control module. S5. The control module adjusts the equipment on the production line in real time based on the test results.
2. The device fault detection method based on the Internet of Things and data-driven approach according to claim 1, characterized in that: The environmental sensors include an environmental humidity sensor, an environmental temperature sensor, and an environmental image recognition sensor; the equipment status sensors include an equipment vibration sensor, an equipment pipeline pressure sensor, an equipment temperature sensor, an equipment image recognition sensor, and a contact displacement sensor.
3. The device fault detection method based on the Internet of Things and data-driven approach according to claim 1, characterized in that: The preprocessing involves sequentially performing A / D conversion, data calibration, and filtering on the data.
4. The device fault detection method based on the Internet of Things and data-driven approach according to claim 3, characterized in that: The A / D conversion includes four steps performed sequentially: sampling, holding, quantization, and encoding.
5. The device fault detection method based on the Internet of Things and data-driven approach according to claim 3, characterized in that: The formula for the filtering algorithm used in the filtering process is set as follows: in, yes The state at any given moment, yes The state at any given moment, It is based on the previous state The prediction results obtained at any time The previous state The optimal result at any given time. Here is the state transition matrix. yes The transpose of the matrix, It is a control input matrix. It is the control input vector. yes The corresponding covariance, yes The corresponding covariance, It is the process noise covariance matrix. For the current moment The optimal estimate; For gain, yes The corresponding covariance, It is the observation matrix. yes The transpose of the matrix, It is the identity matrix. It is the observation noise covariance matrix. yes The actual observation vector at time t.
6. The device fault detection method based on the Internet of Things and data-driven approach according to claim 1, characterized in that: The fault detection module includes a time-frequency processing submodule and a model detection submodule. The time-frequency processing submodule receives data transmitted from the production line environment monitoring module and the equipment status monitoring module, processes the data using a time-frequency analysis algorithm to obtain time-frequency graph data, and transmits the time-frequency graph data to the model detection submodule. The model detection submodule detects the time-frequency graph data to obtain the detection result.
7. The device fault detection method based on the Internet of Things and data-driven approach according to claim 5, characterized in that: The model detection submodule uses a pre-trained convolutional neural network.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.