Industrial safety risk assessment and management system and method based on cognitive computing

By combining deep learning techniques such as convolutional neural networks and long short-term memory networks, the problems of insufficient real-time data processing and low fault prediction accuracy in existing industrial safety risk assessment systems have been solved. This has enabled efficient and adaptive risk management and fault early warning, thereby improving the safety and production efficiency of industrial equipment.

CN118941071AInactive Publication Date: 2024-11-12JIJI SAFETY & ENVIRONMENTAL PROTECTION TECHNOLOGY (CHANGZHOU) CO LTD
View PDF 0 Cites 6 Cited by

Patent Information

Application Number
CN202410673336.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-28
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing industrial safety risk assessment systems suffer from insufficient real-time data processing capabilities, low accuracy in fault prediction, limited adaptive adjustment capabilities, and inadequate handling of long-term data dependencies, resulting in untimely equipment fault response and high safety risks.

Method used

By employing deep learning technology based on cognitive computing, combined with convolutional neural networks and long short-term memory networks, it is integrated into an industrial safety risk assessment and management system. Through data preprocessing, feature learning, pattern recognition, and fault prediction, it generates risk assessment reports and fault warnings.

Benefits of technology

It achieves efficient real-time data processing, improves the accuracy of fault prediction, has adaptive adjustment capabilities, effectively utilizes long-term data, reduces equipment failures and human error, and enhances operational safety and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941071A_ABST
    Figure CN118941071A_ABST
Patent Text Reader

Abstract

The invention discloses an industrial safety risk assessment and management system and method based on cognitive computing, and the method comprises the following steps: S1, collecting operation data from industrial equipment, including equipment states, performance parameters and environmental factors, and carrying out the preprocessing of the collected data; s2, performing feature learning on the preprocessed data by using a convolutional neural network, and identifying a spatial pattern and relevance in the data; s3, taking the output of the convolutional neural network as the input of a long-short term memory network, and analyzing the time sequence features of the data; s4, performing mode identification and fault prediction based on the output of the long and short term memory network in the step S3, and identifying a potential fault mode and an abnormal signal; and S5, generating a risk assessment report and a fault early warning according to the information identified in the step S4, and providing the risk assessment report and the fault early warning to an operator and a manager as decision support. According to the invention, a deep learning technology is fully utilized, and a system for industrial safety risk assessment and management is described.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an industrial safety risk assessment and management system and method based on cognitive computing. Background Art

[0002] In modern industrial production, especially in high-risk industrial environments such as chemical plants, oil refining and heavy machinery manufacturing, the assessment and management of safety risks are crucial. At present, most industrial safety risk assessment and management systems mainly rely on traditional data processing methods and simple machine learning techniques. These systems often face the following core technical difficulties:

[0003] 1. Insufficient real-time data processing capabilities: Traditional systems are usually unable to process and analyze large amounts of data collected from various sensors and devices in real time, resulting in untimely responses to potential failures.

[0004] 2. Low fault prediction accuracy: Due to the lack of in-depth understanding and analysis of complex data patterns, existing systems have low accuracy in fault prediction and cannot effectively identify complex fault patterns.

[0005] 3. Limited adaptive adjustment capabilities: Traditional systems usually lack the ability of self-learning and adaptive adjustment. They cannot automatically optimize the prediction model based on new data and feedback, and are unable to cope with the rapid changes in the industrial environment.

[0006] 4. Insufficient handling of long-term data dependencies: Many existing systems fail to effectively utilize long-term dependencies in historical data, affecting the long-term prediction of equipment performance and the formulation of maintenance plans.

[0007] For example, in traditional industrial equipment fault prediction, the system often requires manual feature selection and fault detection based on simple thresholds or empirical rules, which is not only time-consuming but also prone to errors. In some cases, such as equipment maintenance, conventional methods may involve a multi-step manual inspection process, which is not only inefficient but also may bring safety risks due to improper operation.

[0008] Therefore, how to develop an advanced industrial safety risk assessment and management system that can process large-scale industrial data in real time, accurately predict complex failure modes, and have the ability to self-optimize and adapt to environmental changes is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0009] One object of the present invention is to propose an industrial safety risk assessment and management system and method based on cognitive computing. The present invention makes full use of deep learning technology, especially the integration of convolutional neural networks and long short-term memory networks, and describes in detail a system for industrial safety risk assessment and management.

[0010] An industrial safety risk assessment and management system and method based on cognitive computing according to an embodiment of the present invention includes the following steps:

[0011] S1. Collecting operation data from industrial equipment, including equipment status, performance parameters and environmental factors, and preprocessing the collected data;

[0012] S2, use convolutional neural network to learn features of preprocessed data and identify spatial patterns and correlations in the data;

[0013] S3. Use the output of the convolutional neural network as the input of the long short-term memory network to analyze the time series characteristics of the data;

[0014] S4, based on the output of the long short-term memory network in step S3, implement pattern recognition and fault prediction to identify potential fault modes and abnormal signals;

[0015] S5. Generate a risk assessment report and fault warning based on the information identified in step S4, and provide them to operators and managers as decision support.

[0016] Optionally, the S1 specifically includes:

[0017] S11, collect operation data, the operation data includes device status D s , performance parameters P p and environmental factors f ;

[0018] S12, preprocessing the collected data to form a training set and a test set through data cleaning, normalization and data segmentation;

[0019] S13, forming an input data set D from the processed data set input .

[0020] Optionally, the S2 specifically includes:

[0021] S21. Use convolutional neural networks to learn features from preprocessed data and identify complex spatial patterns and correlations in industrial data through a customized multi-layer convolutional structure.

[0022] S22. Deploy a sequence of convolutional layers designed for industrial data in a convolutional neural network, including optimizing the size, shape, and step size of the convolution kernel, designing multi-scale convolution kernels, and using convolution kernels of different sizes and shapes in each convolution layer. For each layer k, the convolution operation is defined as:

[0023]

[0024] in, This symbol represents the output feature map of the kth convolutional layer. In industrial equipment data analysis, it captures the spatial dependencies of sensor readings from the equipment, temperature distribution, pressure points, and the activation function ReLU, which is used to process nonlinear industrial data. and These are the weights and biases, M k Indicates the number of convolution kernels used in convolution layer k;

[0025] Depthwise separable convolution is applied to enhance the efficiency of feature extraction. Each depthwise separable convolution layer performs depthwise convolution and pointwise convolution successively, and enhances the expressiveness of features by combining the outputs of different layers:

[0026] F sep =σ(W depth ⊙F prev +W point *F depth +d sep )

[0027] where F sep The output of the depthwise separable convolutional layer, W depth ,W point Respectively represent the depth convolution and point-by-point convolution for mining the features of the input channel, ⊙ element-by-element product operation;

[0028] Use the adaptive pooling layer to automatically adjust the pooling window size to adapt to the size of the feature map, thereby balancing the accuracy and processing speed of feature extraction at different scales:

[0029] F pool = adaptive_pool(F sep )

[0030] Among them, F pool and adaptive_pool(F sep ) represent the output of the pooling layer and the adaptive pooling operation respectively;

[0031] S23, by integrating the output of each layer in step S2, a comprehensive feature representation F is formed final ,This representation not only contains the key spatial features of the data, but also optimizes the parsing of the characteristics of industrial equipment through the application of multi-scale and depth-wise separable convolutions, providing high-quality input for the subsequent time series analysis of the long short-term memory network.

[0032] Optionally, the S3 specifically includes:

[0033] S31, receiving the output F of the convolutional neural network final , F finalIt contains comprehensive feature representations extracted from data collected from industrial equipment as input to long short-term memory networks to process time series data;

[0034] S32. Implement a complex gating and memory enhancement mechanism in the LSTM network. The memory enhancement mechanism optimizes the processing effect of time series data by introducing cross-layer feature fusion and time dynamic adjustment. Design a cross-layer feature fusion module. Before each LSTM network unit, use the feature F output by the convolution layer. final Interact with the outputs of the LSTM layers:

[0035] β t =tanh(W β [F final ;h t-1 ]+b β )

[0036] Among them, W β and b β is the learning parameter, β t Used to generate the enhanced feature fusion vector h at time step t t-1 represents the LSTM network output at the previous time step t-1, indicating the system's continuous understanding of the device state;

[0037] Use enhanced feature fusion vector β t For the original input x t Adjust to get the enhanced long short-term memory network input:

[0038] x′ t =β t ⊙x t

[0039] Among them, x t represents the input data at time step t, which are the features F output from the convolutional neural network final , including the spatial characteristics of the industrial equipment status, including the characteristic representation of the temperature, pressure, and vibration sensor data of the equipment;

[0040] For the adjusted input x′ t Perform enhanced LSTM network operations, including custom complex gating operations:

[0041] f t =σ(W f ·[h t-1 ,x′ t ]+U f ·C t-1 +V f ·F final +b f )

[0042] i t =σ(W i ·[h t-1 ,x′ t ]+U i ·C t-1 +V i ·F final +b i )

[0043]

[0044] o t =σ(W o ·[h t-1 ,x′ t ]+U o ·C t +V o ·F final +b o )

[0045] h t =o t *tanh(C t )

[0046] Among them, C t-1 represents the unit state at the previous time step, and stores long-term memory of past information, including historical changes in device performance, α t Attention weight, which represents the weighted input of the current time step, handles device emergencies, σ represents the sigmoid activation function, tanh is the hyperbolic tangent activation function, which is used to adjust the nonlinear characteristics of information flow and help the model handle complex dynamic changes in input data, f t ,i t ,o t are the activation values ​​of the forget gate, input gate, and output gate, respectively. Candidate unit states, W, U these are weight matrices, b bias parameters, representing the linear combination of adjusted gating and activation functions;

[0047] S33. Analyze time series data from industrial equipment through customized gating mechanisms.

[0048] Optionally, the S4 specifically includes:

[0049] S41, receiving the output of the long short-term memory network from step S3, where the output includes processed time series data features for implementing pattern recognition and fault prediction;

[0050] S42. An advanced pattern recognition network is used to analyze the output of the long short-term memory network. The network adopts a multi-layer perceptron structure and uses a fully connected layer to further transform and abstract the output of the long short-term memory network to enhance the model's ability to recognize fault modes:

[0051]

[0052] in, represents the output of the kth fully connected layer, which is used for fault identification when processing complex industrial environment data. prev This is the output of the LSTM network, which contains information about the device state over time, W k and b k are the weights and biases of the kth fully connected layer, α kj This is the weight coefficient for the forward connection;

[0053] After the top fully connected layer, deploy a classification layer:

[0054]

[0055] Among them, F class is the final output of the model, expressed as the probability distribution of the failure mode, β n The coefficient represents the contribution of adjusting the output of each layer in the final fault prediction, W n and b c The weights and biases of the classification layer represent the comprehensive information of the fully connected layer including the fault type. The softmax is normalized and the model output is converted into a probability form. Each classification represents the probability of a specific fault mode.

[0056] Optionally, the S5 specifically includes:

[0057] S51, receiving the fault prediction and pattern recognition results from step S4, which include various potential fault modes and their probability distribution;

[0058] S52, using a decision support module to process the pattern recognition results;

[0059] S53. Generate risk assessment reports and fault warning notifications as visual outputs, including charts and text descriptions, showing failure modes, risk levels, and recommended response measures;

[0060] S54. Provide real-time notifications and reports to operators and managers through the user interface to support rapid decision-making and response.

[0061] Optionally, the S52 specifically includes:

[0062] The decision support module applies a risk scoring model that combines failure probability, failure severity rating, and time sensitivity weighting factors:

[0063]

[0064] Among them, R represents the total risk score, P i represents the probability of the ith failure mode, S i is the severity rating of the corresponding failure mode, t i It represents the time from fault identification to expected fault occurrence, T max The upper limit of the evaluation time window;

[0065] Set up a dynamic warning level algorithm based on risk scoring:

[0066]

[0067] Among them, T h and T l They represent the thresholds for high risk and low risk respectively, δR is the change in the current risk score compared with the previous scoring cycle, and log(1+δR) is the risk change sensitivity function, which is used to adjust the risk threshold to make the warning level more sensitive to rapid changes in risk.

[0068] Optionally, an industrial safety risk assessment and management system based on cognitive computing includes the following modules:

[0069] A data collection unit, for collecting operation data from industrial equipment, including equipment status, performance parameters and environmental factors, and preprocessing the collected data, including data cleaning, normalization and feature extraction;

[0070] A deep learning processing unit, configured with a convolutional neural network and a long short-term memory network, where the convolutional neural network is used to extract spatial features from preprocessed data, and the long short-term memory network is used to analyze time series data to achieve complex fault pattern recognition and prediction;

[0071] Pattern recognition and risk assessment unit, including multi-layer perceptron and decision support module. The multi-layer perceptron is used to further analyze the characteristics of the output of the long short-term memory network. The decision support module applies complex logical rules and risk scoring models to generate risk assessment reports and fault warnings.

[0072] Adaptive learning unit, which automatically adjusts the model parameters of convolutional neural network and long short-term memory network according to newly collected data and operational feedback;

[0073] User interface unit for displaying real-time notifications and risk assessment reports to operators and managers, supporting decision making and rapid implementation of response measures, clearly displaying failure modes, risk levels and recommended response measures through diagrams and text descriptions;

[0074] The communication interface unit is responsible for data exchange between the system and industrial equipment, and communication with external systems, supporting real-time updating and processing of data;

[0075] The control strategy unit based on cognitive computing implements automated control strategies based on data analysis results and adjusts equipment operation and maintenance plans in real time.

[0076] The beneficial effects of the present invention are:

[0077] The present invention makes full use of deep learning technology, especially the integration of convolutional neural networks and long short-term memory networks, and describes in detail a system for industrial safety risk assessment and management. The system has the following beneficial effects:

[0078] 1. Efficient real-time data processing: By integrating convolutional neural networks and recurrent neural networks, this system is able to process and analyze large amounts of data collected from industrial equipment in real time. This real-time processing capability ensures that potential faults and safety risks can be identified and handled immediately, greatly reducing the risk of equipment failure and production interruption.

[0079] 2. Improve the accuracy of fault prediction: The system uses advanced deep learning models to identify complex fault modes. These models can understand and analyze complex patterns in data, thereby providing more accurate fault predictions than traditional methods. This increase in accuracy helps companies to conduct risk management and maintenance planning in advance, thereby optimizing operational efficiency and costs.

[0080] 3. Adaptive adjustment and continuous learning: The system has the ability to self-optimize and automatically adjust its prediction model based on new data and feedback. This ability to adapt and continuously learn enables the system to cope with rapid changes in the industrial environment and ensure the continuity and accuracy of its prediction performance.

[0081] 4. Effective use of long-term data dependencies: Through the long short-term memory network, the system can analyze and learn long-term data dependencies, which not only enhances the accuracy of fault prediction, but also provides data support for formulating more accurate maintenance plans.

[0082] 5. Improve operational safety and reduce manpower requirements: The system reduces reliance on on-site personnel through automated risk assessment and fault warning, thereby reducing safety hazards for personnel working in high-risk environments. In addition, intelligent fault diagnosis and processing reduces human errors and improves the safety and efficiency of overall operations.

[0083] Therefore, the present invention provides solid technical support for replacing traditional manual risk management methods with machine learning and cognitive computing technologies, making industrial safety management more intelligent, efficient and safe. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0085] Figure 1 The overall architecture diagram of the deep learning-based industrial safety risk assessment and management system implemented by the present invention;

[0086] Figure 2 The data flow diagram of the data collection unit is shown, illustrating the process of collecting and preprocessing operating data from industrial equipment. DETAILED DESCRIPTION

[0087] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0088] refer to Figure 1 , an industrial safety risk assessment and management system and method based on cognitive computing, comprising the following steps:

[0089] S1. Collecting operation data from industrial equipment, including equipment status, performance parameters and environmental factors, and preprocessing the collected data;

[0090] In this implementation, S1 specifically includes:

[0091] S11, collect operation data, the operation data includes device status D s , performance parameters P p and environmental factors f ;

[0092] S12, preprocessing the collected data to form a training set and a test set through data cleaning, normalization and data segmentation;

[0093] S13, forming an input data set D from the processed data set input .

[0094] S2, use convolutional neural network to learn features of preprocessed data and identify spatial patterns and correlations in the data;

[0095] In this implementation, S2 specifically includes:

[0096] S21. Use convolutional neural networks to learn features from preprocessed data and identify complex spatial patterns and correlations in industrial data through a customized multi-layer convolutional structure.

[0097] S22. Deploy a sequence of convolutional layers designed for industrial data in a convolutional neural network, including optimizing the size, shape, and step size of the convolution kernel, designing multi-scale convolution kernels, and using convolution kernels of different sizes and shapes in each convolution layer. For each layer k, the convolution operation is defined as:

[0098]

[0099] in, This symbol represents the output feature map of the kth convolutional layer. In industrial equipment data analysis, it captures the spatial dependencies of sensor readings from the equipment, temperature distribution, pressure points, and the activation function ReLU, which is used to process nonlinear industrial data. and These are the weights and biases, M k Indicates the number of convolution kernels used in convolution layer k;

[0100] Depthwise separable convolution is applied to enhance the efficiency of feature extraction. Each depthwise separable convolution layer performs depthwise convolution and pointwise convolution successively, and enhances the expressiveness of features by combining the outputs of different layers:

[0101] F sep =σ(W depth ⊙F prev +W point *F depth +b sep )

[0102] Among them, F sep The output of the depthwise separable convolutional layer, W depth ,W point Respectively represent the depth convolution and point-by-point convolution for mining the features of the input channel, ⊙ element-by-element product operation;

[0103] Use the adaptive pooling layer to automatically adjust the pooling window size to adapt to the size of the feature map, thereby balancing the accuracy and processing speed of feature extraction at different scales:

[0104] F pool = adaptive_pool(F sep )

[0105] Among them, F pool and adaptive_pool(F sep ) represent the output of the pooling layer and the adaptive pooling operation respectively;

[0106] S23, by integrating the output of each layer in step S2, a comprehensive feature representation F is formed final ,This representation not only contains the key spatial features of the data, but also optimizes the parsing of the characteristics of industrial equipment through the application of multi-scale and depth-wise separable convolutions, providing high-quality input for the subsequent time series analysis of the long short-term memory network.

[0107] S3. Use the output of the convolutional neural network as the input of the long short-term memory network to analyze the time series characteristics of the data;

[0108] In this implementation, S3 specifically includes:

[0109] S31, receiving the output F of the convolutional neural network final , F final It contains comprehensive feature representations extracted from data collected from industrial equipment as input to long short-term memory networks to process time series data;

[0110] S32. Implement a complex gating and memory enhancement mechanism in the LSTM network. The memory enhancement mechanism optimizes the processing effect of time series data by introducing cross-layer feature fusion and time dynamic adjustment. Design a cross-layer feature fusion module. Before each LSTM network unit, use the feature F output by the convolution layer. final Interact with the outputs of the LSTM layers:

[0111] β t =tanh(W β [F final ;h t-1 ]+b β )

[0112] Among them, W β and b β is the learning parameter, β t Used to generate the enhanced feature fusion vector h at time step t t-1 represents the LSTM network output at the previous time step t-1, indicating the system's continuous understanding of the device state;

[0113] Use enhanced feature fusion vector β t For the original input x t Adjust to get the enhanced long short-term memory network input:

[0114] x′ t =β t ⊙x t

[0115] Among them, x t represents the input data at time step t, which are the features F output from the convolutional neural networkfinal , including the spatial characteristics of the industrial equipment status, including the characteristic representation of the temperature, pressure, and vibration sensor data of the equipment;

[0116] For the adjusted input x′ t Perform enhanced LSTM network operations, including custom complex gating operations:

[0117] f t =σ(W f ·[h t-1 ,x′ t ]+U f ·C t-1 +V f ·F final +b f )

[0118] i t =σ(W i ·[h t-1 ,x′ t ]+U i ·C t-1 +V i ·F final +b i )

[0119]

[0120] o t =σ(W o ·[h t-1 ,x′ t ]+U o ·C t +V o ·F final +b o )

[0121] h t =o t *tanh(C t )

[0122] Among them, C t-1 represents the unit state at the previous time step, and stores long-term memory of past information, including historical changes in device performance, α t Attention weight, which represents the weighted input of the current time step, handles device emergencies, σ represents the sigmoid activation function, tanh is the hyperbolic tangent activation function, which is used to adjust the nonlinear characteristics of information flow and help the model handle complex dynamic changes in input data, f t ,i t ,o t are the activation values ​​of the forget gate, input gate, and output gate, respectively. Candidate unit states, W, U these are weight matrices, b bias parameters, representing the linear combination of adjusted gating and activation functions;

[0123] S33. Analyze time series data from industrial equipment through customized gating mechanisms.

[0124] S4, based on the output of the long short-term memory network in step S3, implement pattern recognition and fault prediction to identify potential fault modes and abnormal signals;

[0125] In this implementation, S4 specifically includes:

[0126] S41, receiving the output of the long short-term memory network from step S3, where the output includes processed time series data features for implementing pattern recognition and fault prediction;

[0127] S42. An advanced pattern recognition network is used to analyze the output of the long short-term memory network. The network adopts a multi-layer perceptron structure and uses a fully connected layer to further transform and abstract the output of the long short-term memory network to enhance the model's ability to recognize fault modes:

[0128]

[0129] in, represents the output of the kth fully connected layer, which is used for fault identification when processing complex industrial environment data. prev This is the output of the LSTM network, which contains information about the device state over time, W k and b k are the weights and biases of the kth fully connected layer, α kj This is the weight coefficient for the forward connection;

[0130] After the top fully connected layer, deploy a classification layer:

[0131]

[0132] Among them, F class is the final output of the model, expressed as the probability distribution of the failure mode, β n The coefficient represents the contribution of adjusting the output of each layer in the final fault prediction, W n and b c The weights and biases of the classification layer represent the comprehensive information of the fully connected layer including the fault type. The softmax is normalized and the model output is converted into a probability form. Each classification represents the probability of a specific fault mode.

[0133] S5. Generate a risk assessment report and fault warning based on the information identified in step S4, and provide them to operators and managers as decision support.

[0134] In this implementation, S5 specifically includes:

[0135] S51, receiving the fault prediction and pattern recognition results from step S4, which include various potential fault modes and their probability distribution;

[0136] S52, using a decision support module to process the pattern recognition results;

[0137] S53. Generate risk assessment reports and fault warning notifications as visual outputs, including charts and text descriptions, showing failure modes, risk levels, and recommended response measures;

[0138] S54. Provide real-time notifications and reports to operators and managers through the user interface to support rapid decision-making and response.

[0139] S52, using a decision support module to process the pattern recognition results;

[0140] In this implementation manner, S52 specifically includes:

[0141] The decision support module applies a risk scoring model that combines failure probability, failure severity rating, and time sensitivity weighting factors:

[0142]

[0143] Among them, R represents the total risk score, P i represents the probability of the ith failure mode, S i is the severity rating of the corresponding failure mode, t i It represents the time from fault identification to expected fault occurrence, T max The upper limit of the evaluation time window;

[0144] Set up a dynamic warning level algorithm based on risk scoring:

[0145]

[0146] Among them, T h and T l They represent the thresholds for high risk and low risk respectively, δR is the change in the current risk score compared with the previous scoring cycle, and log(1+δR) is the risk change sensitivity function, which is used to adjust the risk threshold to make the warning level more sensitive to rapid changes in risk.

[0147] In this embodiment, an industrial safety risk assessment and management system based on cognitive computing includes the following modules:

[0148] A data collection unit, for collecting operation data from industrial equipment, including equipment status, performance parameters and environmental factors, and preprocessing the collected data, including data cleaning, normalization and feature extraction;

[0149] A deep learning processing unit, configured with a convolutional neural network and a long short-term memory network, where the convolutional neural network is used to extract spatial features from preprocessed data, and the long short-term memory network is used to analyze time series data to achieve complex fault pattern recognition and prediction;

[0150] Pattern recognition and risk assessment unit, including multi-layer perceptron and decision support module. The multi-layer perceptron is used to further analyze the characteristics of the output of the long short-term memory network. The decision support module applies complex logical rules and risk scoring models to generate risk assessment reports and fault warnings.

[0151] Adaptive learning unit, which automatically adjusts the model parameters of convolutional neural network and long short-term memory network according to newly collected data and operational feedback;

[0152] User interface unit for displaying real-time notifications and risk assessment reports to operators and managers, supporting decision making and rapid implementation of response measures, clearly displaying failure modes, risk levels and recommended response measures through diagrams and text descriptions;

[0153] The communication interface unit is responsible for data exchange between the system and industrial equipment, and communication with external systems, supporting real-time updating and processing of data;

[0154] The control strategy unit based on cognitive computing implements automated control strategies based on data analysis results and adjusts equipment operation and maintenance plans in real time.

[0155] Embodiment 1:

[0156] This example describes in detail the application of a deep learning-based industrial safety risk assessment and management system in a large chemical plant in Jiangsu Province, China in 2023. The chemical plant mainly produces high-performance polymers and chemical solvents, with 5 main production lines, each equipped with multiple reactors, distillation towers and large-capacity storage containers. Previously, the plant had frequent accidents caused by equipment failure, especially a pressure tank leakage accident that was not discovered in time, which caused direct economic losses of more than one million yuan and serious pollution to the surrounding environment.

[0157] Scenario Description

[0158] Before implementing the new system, the chemical plant relied on traditional sensor monitoring and periodic manual inspections to manage operational risks. In March 2023, management decided to adopt the system provided by the present invention to comprehensively improve the level of risk management. After the system is deployed, it is connected to the sensors of all key equipment to monitor key indicators including temperature, pressure, pH value, solvent concentration, etc. in real time.

[0159] System Application

[0160] The system's data collection unit collects data from all device sensors in real time, and then processes it through the preprocessing unit, including denoising, normalization, and extraction of key indicators. Through convolutional neural networks, the system analyzes the spatial characteristics of the data, such as uneven temperature distribution and pressure fluctuations. Subsequently, the long short-term memory network analyzes the evolution of these spatial characteristics over time and captures potential abnormal patterns.

[0161] In an incident in June 2023, the system successfully predicted an abnormal increase in pressure in a distillation tower. The pressure change trend captured by the long short-term memory network model gradually deviated from the normal working range, and the pattern recognition and risk assessment unit predicted the potential risk of equipment failure based on this.

[0162] Table 1 Comparison of system performance of Jiangsu chemical plant

[0163]

[0164] Referring to Table 1 above, during the six-month implementation period from March 2023 to September 2023, the system monitored and analyzed more than 50 potential faults and abnormal events. The timeliness of fault detection has been advanced from the previous average of 6 hours to within 15 minutes, and the accuracy has been improved from about 75% in the past to 98%. In a critical reactor overheating event that occurred in September, the system's warning time was 18 hours ahead of the actual event, an increase of 150% over the previous average time.

[0165] in conclusion

[0166] Through the implementation of the system of the present invention, Jiangsu Chemical Plant has significantly improved its monitoring and risk management capabilities for industrial equipment, reducing the number of accidents and potential economic losses caused by failures not discovered in time. In addition, the self-learning and adaptive adjustment functions enable the system to self-optimize after each event analysis to adapt to the changing conditions on the production line, thereby achieving a more efficient and safer production process.

[0167] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An industrial safety risk assessment and management system and method based on cognitive computing, characterized in that: The steps include: S1. Collecting operation data from industrial equipment, including equipment status, performance parameters and environmental factors, and preprocessing the collected data; S2, use convolutional neural network to learn features of preprocessed data and identify spatial patterns and correlations in the data; S3. Use the output of the convolutional neural network as the input of the long short-term memory network to analyze the time series characteristics of the data; S4, based on the output of the long short-term memory network in step S3, implement pattern recognition and fault prediction to identify potential fault modes and abnormal signals; S5. Generate a risk assessment report and fault warning based on the information identified in step S4, and provide them to operators and managers as decision support.

2. The industrial safety risk assessment and management system and method based on cognitive computing according to claim 1 is characterized in that: The S1 specifically includes: S11, collect operation data, the operation data includes device status D s , performance parameters P p and environmental factors f ; S12, preprocessing the collected data to form a training set and a test set through data cleaning, normalization and data segmentation; S13, forming an input data set D from the processed data set input .

3. The industrial safety risk assessment and management system and method based on cognitive computing according to claim 1 is characterized in that: The S2 specifically includes: S21. Use convolutional neural networks to learn features from preprocessed data and identify complex spatial patterns and correlations in industrial data through a customized multi-layer convolutional structure. S22. Deploy a sequence of convolutional layers designed for industrial data in a convolutional neural network, including optimizing the size, shape, and step size of the convolution kernel, designing multi-scale convolution kernels, and using convolution kernels of different sizes and shapes in each convolution layer. For each layer k, the convolution operation is defined as: in, This symbol represents the output feature map of the kth convolutional layer. In industrial equipment data analysis, it captures the spatial dependencies of sensor readings from the equipment, temperature distribution, pressure points, and the activation function ReLU, which is used to process nonlinear industrial data. and These are the weights and biases, M k Indicates the number of convolution kernels used in convolution layer k; Depthwise separable convolution is applied to enhance the efficiency of feature extraction. Each depthwise separable convolution layer performs depthwise convolution and pointwise convolution successively, and enhances the expressiveness of features by combining the outputs of different layers: F sep =σ(W depth ⊙F prev +W point *F depth +d sep ) where F sep The output of the depthwise separable convolutional layer, W depth ,W point Respectively represent the depth convolution and point-by-point convolution for mining the features of the input channel, ⊙ element-by-element product operation; Use the adaptive pooling layer to automatically adjust the pooling window size to adapt to the size of the feature map, thereby balancing the accuracy and processing speed of feature extraction at different scales: F pool =adaptive_pool(F sep ) Among them, F pool and adaptive_pool(F sep ) represent the output of the pooling layer and the adaptive pooling operation respectively; S23, by integrating the output of each layer in step S2, a comprehensive feature representation F is formed final ,This representation not only contains the key spatial features of the data, but also optimizes the parsing of the characteristics of industrial equipment through the application of multi-scale and depth-wise separable convolutions, providing high-quality input for the subsequent time series analysis of the long short-term memory network.

4. The industrial safety risk assessment and management system and method based on cognitive computing according to claim 1 is characterized in that: The S3 specifically includes: S31, receiving the output F of the convolutional neural network final , F final It contains comprehensive feature representations extracted from data collected from industrial equipment as input to long short-term memory networks to process time series data; S32. Implement a complex gating and memory enhancement mechanism in the LSTM network. The memory enhancement mechanism optimizes the processing effect of time series data by introducing cross-layer feature fusion and time dynamic adjustment. Design a cross-layer feature fusion module. Before each LSTM network unit, use the feature F output by the convolution layer. final Interact with the outputs of the LSTM layers: β t =tanh(W β [F final ;h t-1 ]+b β ) Among them, W β and b β is the learning parameter, β t Used to generate the enhanced feature fusion vector h at time step t t-1 represents the LSTM network output at the previous time step t-1, indicating the system's continuous understanding of the device state; Use enhanced feature fusion vector β t For the original input x t Adjust to get the enhanced long short-term memory network input: x′ t =b t ⊙x t Among them, x t represents the input data at time step t, which are the features F output from the convolutional neural network final , including the spatial characteristics of the industrial equipment status, including the characteristic representation of the temperature, pressure, and vibration sensor data of the equipment; For the adjusted input x′ t Perform enhanced LSTM network operations, including custom complex gating operations: f t =σ(W f ·[h t-1 ,x′ t ]+U f ·C t-1 +V f ·F final +b f ) i t =σ(W i ·[h t-1 ,x′ t ]+U i ·C t-1 +V i ·F final +b i ) o t =σ(W o ·[h t-1 ,x′ t ]+U o ·C t +V o ·F final +b o ) h t =o t *tanh(C t ) Among them, C t-1 represents the unit state at the previous time step, and stores long-term memory of past information, including historical changes in device performance, α t Attention weight, which represents the weighted input of the current time step, handles device emergencies, σ represents the sigmoid activation function, tanh is the hyperbolic tangent activation function, which is used to adjust the nonlinear characteristics of information flow and help the model handle complex dynamic changes in input data, f t ,i t ,o t are the activation values ​​of the forget gate, input gate, and output gate, respectively. Candidate unit states, W, U these are weight matrices, b bias parameters, representing the linear combination of adjusted gating and activation functions; S33. Analyze time series data from industrial equipment through customized gating mechanisms.

5. The industrial safety risk assessment and management system and method based on cognitive computing according to claim 1 is characterized in that: The S4 specifically includes: S41, receiving the output of the long short-term memory network from step S3, where the output includes processed time series data features for implementing pattern recognition and fault prediction; S42. An advanced pattern recognition network is used to analyze the output of the long short-term memory network. The network adopts a multi-layer perceptron structure and uses a fully connected layer to further transform and abstract the output of the long short-term memory network to enhance the model's ability to recognize fault modes: in, represents the output of the kth fully connected layer, which is used for fault identification when processing complex industrial environment data. prev This is the output of the LSTM network, which contains information about the device state over time, W k and b k are the weights and biases of the kth fully connected layer, α kj This is the weight coefficient for the forward connection; After the top fully connected layer, deploy a classification layer: Among them, F class is the final output of the model, expressed as the probability distribution of the failure mode, β n The coefficient represents the contribution of adjusting the output of each layer in the final fault prediction, W n and b c The weights and biases of the classification layer represent the comprehensive information of the fully connected layer including the fault type. The softmax is normalized and the model output is converted into a probability form. Each classification represents the probability of a specific fault mode.

6. The industrial safety risk assessment and management system and method based on cognitive computing according to claim 1 is characterized in that: The S5 specifically includes: S51, receiving the fault prediction and pattern recognition results from step S4, which include various potential fault modes and their probability distribution; S52, using a decision support module to process the pattern recognition results; S53. Generate risk assessment reports and fault warning notifications as visual outputs, including charts and text descriptions, showing failure modes, risk levels, and recommended response measures; S54. Provide real-time notifications and reports to operators and managers through the user interface to support rapid decision-making and response.

7. The industrial safety risk assessment and management system and method based on cognitive computing according to claim 1 is characterized in that: The S52 specifically includes: The decision support module applies a risk scoring model that combines failure probability, failure severity rating, and time sensitivity weighting factors: Among them, R represents the total risk score, P i represents the probability of the ith failure mode, S i is the severity rating of the corresponding failure mode, t i It represents the time from fault identification to expected fault occurrence, T max The upper limit of the evaluation time window; Set up a dynamic warning level algorithm based on risk scoring: Among them, T h and T l They represent the thresholds for high risk and low risk respectively, δR is the change in the current risk score compared with the previous scoring cycle, and log(1+δR) is the risk change sensitivity function, which is used to adjust the risk threshold to make the warning level more sensitive to rapid changes in risk.

8. An industrial safety risk assessment and management system based on cognitive computing, characterized in that: Includes the following modules: A data collection unit, for collecting operation data from industrial equipment, including equipment status, performance parameters and environmental factors, and preprocessing the collected data, including data cleaning, normalization and feature extraction; A deep learning processing unit, configured with a convolutional neural network and a long short-term memory network, where the convolutional neural network is used to extract spatial features from preprocessed data, and the long short-term memory network is used to analyze time series data to achieve complex fault pattern recognition and prediction; Pattern recognition and risk assessment unit, including multi-layer perceptron and decision support module. The multi-layer perceptron is used to further analyze the characteristics of the output of the long short-term memory network. The decision support module applies complex logical rules and risk scoring models to generate risk assessment reports and fault warnings. Adaptive learning unit, which automatically adjusts the model parameters of convolutional neural network and long short-term memory network according to newly collected data and operational feedback; User interface unit for displaying real-time notifications and risk assessment reports to operators and managers, supporting decision making and rapid implementation of response measures, clearly displaying failure modes, risk levels and recommended response measures through diagrams and text descriptions; The communication interface unit is responsible for data exchange between the system and industrial equipment, and communication with external systems, supporting real-time updating and processing of data; The control strategy unit based on cognitive computing implements automated control strategies based on data analysis results and adjusts equipment operation and maintenance plans in real time.

Citation Information

Cited By

  • Network data leakage monitoring system and method

    CN119561794A

  • Network Data Leakage Monitoring System and Method

    CN119561794B

  • Building information processing method based on Internet of Things

    CN119991354A

  • Safety management method for comprehensive security and protection integrated management system

    CN120070135A

  • Security management method for integrated security integrated management system

    CN120070135B