Chemical engineering design and safety risk analysis integrated platform device

Abstract

The present application relates to the technical field of industrial safety and chemical engineering design, in particular to a chemical engineering design and safety risk analysis integrated platform.The platform adopts a four-layer architecture of data base layer, intelligent engine layer, business component layer and human-computer interaction layer.The method comprises the following steps: intelligently analyzing chemical engineering drawings through an AI-P&ID recognition engine, extracting drawing element parameters and constructing topological relations; performing parameterized editing and conflict checking through a P&ID editing and checking component; performing process flow dynamic simulation through a process simulation engine; performing intelligent reasoning of risks through a HAZOP knowledge base large model; and finally automatically generating a structured risk analysis report through a report generation component.The present application realizes full-chain automation from drawing analysis to report generation, solves the problems of serious manpower dependence, obvious data defects, low efficiency and fragmented functions in traditional methods, improves the efficiency, accuracy and reliability of chemical safety risk analysis, and reduces the cost of manpower and software.

Description

Technical Field

[0001] This invention relates to the field of industrial safety and chemical engineering design technology, and more specifically to an integrated platform for chemical design and safety risk analysis. Background Technology

[0002] In the existing technology, chemical engineering design and safety risk analysis mainly rely on commercial software or publicly available patented technology solutions, but these solutions have many limitations.

[0003] Commercial software such as AVEVA E3D supports P&ID drawing and HAZOP analysis, but requires manual import of drawings. Its risk reasoning function relies on a preset rule base, lacks self-evolution capabilities, and requires a large number of manual intervention steps.

[0004] Another commercial software, Intergraph Smart Plant Hazard Analysis, needs to be used in conjunction with CAD software. It does not have independent AI recognition capabilities and cannot support real-time data integration and dynamic simulation, resulting in fragmented functionality.

[0005] Regarding publicly disclosed patent technologies, for example, the chemical HAZOP analysis method described in CN114547818A uses an expert system for reasoning, but does not integrate AI image recognition functions. Parameter input relies on manual operation and lacks a dynamic simulation step. On the other hand, the P&ID recognition method based on deep learning described in CN115081857B can only realize the primitive detection function, does not integrate HAZOP reasoning and process simulation capabilities, and cannot form a closed-loop processing chain.

[0006] Based on the aforementioned existing technologies, the technical problems can be summarized as follows:

[0007] First, it is heavily reliant on human resources. Traditional HAZOP analysis requires collaboration among multidisciplinary teams. Incomplete staffing or communication barriers across disciplines can easily lead to low analysis efficiency and biased conclusions. Furthermore, processes such as drawing processing, parameter input, risk analysis, and report writing are all highly dependent on manual operation, which is not only time-consuming but also prone to introducing human error.

[0008] Secondly, the data defects are obvious. The drawings of old equipment are often missing or faded, which is difficult to repair manually. At the same time, the data of different departments are scattered and there are significant obstacles to sharing, resulting in incomplete analysis data and affecting the accuracy of the conclusions.

[0009] Secondly, there is an imbalance between efficiency and cost. The analysis cycle usually takes several days to a week. The labor cost and the purchase cost of using multiple single-function software are high. Furthermore, incomplete data or human error can easily lead to design rework, which further increases the overall cost.

[0010] Finally, the problem of functional fragmentation is prominent. Existing software focuses on a single function, such as only realizing P&ID identification or only performing risk analysis. It requires the linkage and cooperation of multiple software programs, and data transmission relies on manual operation. It cannot achieve full-chain automation from design to simulation to analysis.

[0011] The aforementioned technical issues collectively restrict the efficient, accurate, and low-cost implementation of chemical engineering design and safety risk analysis.

[0012] With the increasing demands for process safety management in the chemical industry, hazard and operability analysis (HAZOP) has become a crucial link in ensuring production safety, leading to an increasingly urgent need for efficient and accurate HAZOP analysis tools. Simultaneously, standardized output of HAZOP analysis reports has become a common industry requirement, necessitating the support of corresponding software platforms. Summary of the Invention

[0013] In view of this, the present invention provides an integrated platform for chemical engineering design and safety risk analysis, which aims to solve the problems of broken analysis chains caused by functional fragmentation, low efficiency and subjective errors caused by high dependence on human intervention, incomplete risk identification coverage caused by data defects and rigid rules, and high cost and operational complexity caused by multi-software collaboration in the prior art, thereby improving the automation level, analysis efficiency, reliability of conclusions and overall economy of the chemical safety risk analysis process.

[0014] To achieve the above objectives, the present invention adopts the following technical solution:

[0015] An integrated platform for chemical engineering design and safety risk analysis, the platform comprising:

[0016] The data foundation layer is used to integrate multi-source data and perform data cleaning and structure transformation;

[0017] The intelligent engine layer, which is communicatively connected to the data base layer, is used to execute core algorithm processing, including the AI-P&ID recognition engine, the process simulation engine, and the HAZOP knowledge base large model.

[0018] The business component layer, which communicates with the intelligent engine layer, is used to call the core algorithm to process the results to realize core business functions. It includes the P&ID intelligent parsing component, the P&ID editing and verification component, the dynamic simulation analysis component, the HAZOP risk reasoning component, and the report generation component.

[0019] The human-computer interaction layer communicates with the business component layer and is used to provide a user interface and support interface with external systems.

[0020] In one specific implementation scheme, the data base layer uses distributed database technology to realize data storage and governance.

[0021] In one specific implementation scheme, the AI-P&ID recognition engine uses a joint model built on an improved YOLO algorithm and a Diffusion generative model to achieve primitive detection and low-quality drawing repair of chemical engineering drawings.

[0022] In one specific implementation, the improved YOLO algorithm incorporates a CBAM attention module, a bidirectional cross-fusion feature fusion path, and a CIoU loss function.

[0023] In one specific implementation, the process simulation engine is configured to perform steady-state simulations based on the sequential module method and dynamic simulations based on the finite difference method.

[0024] In one specific implementation scheme, the HAZOP knowledge base model is built on the Transformer architecture and integrates risk cases and a guide vocabulary that have undergone data cleaning and feature extraction.

[0025] In one specific implementation, the P&ID editing and verification component integrates a process rule library and is configured with a reinforcement learning-based rule self-evolution mechanism for dynamically optimizing conflict verification rules.

[0026] In one specific implementation, the report generation component is configured to integrate risk reasoning results, process simulation parameters, and P&ID drawing basic information, and combine natural language generation technology with a visualization engine to output a structured report.

[0027] In one specific implementation, the platform is configured to perform the following automated processes:

[0028] The AI-P&ID recognition engine intelligently analyzes chemical engineering drawings, extracts primitive parameters, and constructs topological relationships.

[0029] The P&ID editing and verification component performs parametric editing and conflict verification on the parsed drawing data;

[0030] The process simulation engine performs dynamic simulation of the process flow based on the verified data;

[0031] The HAZOP risk reasoning component performs intelligent risk reasoning based on simulation results;

[0032] The report generation component automatically generates a structured risk analysis report.

[0033] In one specific implementation scheme, the AI-P&ID recognition engine performs intelligent parsing of chemical engineering drawings by: using a joint model based on an improved YOLO algorithm and a Diffusion generation model to perform integrated processing of low-quality drawing repair and primitive detection; the P&ID editing and verification component performs conflict verification by implementing a self-evolutionary process of dynamic rule optimization based on a reinforcement learning agent.

[0034] Compared with existing technologies, the integrated platform for chemical engineering design and safety risk analysis described in this invention is used to achieve integrated processing of chemical engineering design and safety risk analysis. By constructing a software system architecture that includes a data foundation layer, an intelligent engine layer, a business component layer, and a human-computer interaction layer, and combining core algorithms such as the AI-P&ID recognition engine, the process simulation engine, and the HAZOP knowledge base model, it achieves a fully automated closed loop from intelligent drawing parsing, parametric editing and verification, dynamic process simulation, to intelligent risk reasoning and automatic generation of structured reports. This effectively improves analysis efficiency, reduces costs, and enhances the reliability of conclusions, offering the following beneficial effects:

[0035] First, efficiency improvement

[0036] The AI-P&ID recognition engine automates the parsing of chemical drawings, replacing manual drawing annotation; the HAZOP knowledge base model automates risk reasoning, replacing manual risk analysis; and the automatic report generation module automates report writing, replacing manual report preparation. Meanwhile, the interface linkage between various software modules enables automatic data flow, avoiding manual data transmission, thereby significantly shortening the HAZOP analysis cycle and accelerating the risk warning response speed.

[0037] Second, cost reduction

[0038] By utilizing automated drawing parsing technology, we reduced human resource input. Through rule self-evolution and virtual-real verification closed loop, we reduced the design rework rate caused by human error. Furthermore, by integrating multiple functions through an integrated platform, we replaced the use of multiple single-function software programs. As a result, we achieved overall cost optimization by reducing human resource input, avoiding rework losses, and saving software purchase costs.

[0039] Third, improved accuracy

[0040] By leveraging the self-learning ability of topological relationships, implicit topological risks in drawings are identified. The compliance and on-site adaptability of design schemes are improved through the rule self-evolution mechanism and virtual-real verification. The HAZOP knowledge base model based on the Transformer architecture is used to perform reasoning based on massive risk cases and standardized rules to reduce human subjective bias. Furthermore, dynamic simulation technology is combined to identify dynamic working condition risks that are difficult to detect by traditional static analysis, thereby comprehensively improving the coverage of risk identification and the rigor of conclusions.

[0041] Fourth, enhanced applicability

[0042] The low-quality drawing repair function based on the Diffusion-YOLO joint model can effectively handle the problems of fading and creases in drawings of old equipment. Through transfer learning, the model is fine-tuned to adapt to the symbol habits of different enterprises. The rule base integration mechanism supports the import of enterprise-customized rules. At the same time, the system docking capability provided by the human-computer interaction layer ensures data interoperability with the enterprise's existing system, so that the platform can adapt to diverse enterprise application scenarios and data conditions. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the modules and data flow of an integrated platform for chemical engineering design and safety risk analysis as described in this invention. Detailed Implementation

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] This invention provides an integrated platform for chemical engineering design and safety risk analysis. By integrating AI technology (including image recognition and large model inference), process simulation and module collaboration, it aims to solve the problems of heavy reliance on human labor, obvious data defects, low analysis efficiency and functional fragmentation in existing technologies. It achieves full-chain automation from "drawing analysis - design verification - process simulation - risk inference - report generation" to meet the industry's urgent need for efficient, accurate and low-cost HAZOP analysis.

[0047] To achieve the above objectives, this platform fundamentally solves the problem of fragmented analysis chains caused by functional fragmentation by constructing an integrated software system architecture. Specifically, the platform includes a data foundation layer, an intelligent engine layer, a business component layer, and a human-computer interaction layer. The data foundation layer integrates and manages multi-source data; the intelligent engine layer deploys an AI-P&ID recognition engine, a process simulation engine, and a large HAZOP knowledge base model, responsible for executing core algorithm processing; the business component layer contains multiple collaborative functional modules used to realize the entire chain of business functions, including drawing parsing, parameter editing and verification, process simulation, risk reasoning, and report generation; and the human-computer interaction layer provides a user interface and the ability to interface with external systems. Each software layer interacts with data and transmits instructions through standardized interfaces, thereby achieving an automated closed loop for chemical engineering design and safety risk analysis within a single platform. This effectively reduces reliance on manual labor and switching between multiple software programs, improving analysis efficiency and the reliability of conclusions.

[0048] like Figure 1 As shown, the integrated platform for chemical engineering design and safety risk analysis described in this invention adopts a layered software architecture, including a data foundation layer, an intelligent engine layer, a business component layer, and a human-computer interaction layer. Each layer interacts with the others through standardized interfaces for data exchange and command transmission, collaborating to complete the entire automated process of chemical engineering design and safety risk analysis. Its specific working process is as follows:

[0049] First, in the intelligent parsing stage of chemical engineering drawings, users upload P&ID drawings through the WEB interface of the human-computer interaction layer. This command triggers the P&ID intelligent parsing component in the business component layer, which then calls the AI-P&ID recognition engine in the intelligent engine layer. This engine performs the drawing parsing task based on a joint model built on an improved YOLO algorithm and a Diffusion generative model. Specifically, the improved YOLO algorithm improves primitive detection accuracy by embedding a CBAM attention module, adopting a bidirectional cross-fusion BiFPN feature path, and a CIoU loss function; the Diffusion model repairs low-quality drawings through an iterative "denoising-generation" process that includes noise injection, noise prediction, and detail restoration. The engine then performs image preprocessing, primitive detection, and parameter extraction, and constructs a "equipment-pipeline-instrument" topological relationship based on a graph neural network (GNN), finally returning the parsed primitive attribute library.

[0050] Subsequently, the P&ID parameterization editing and conflict verification phase begins. The P&ID editing and verification component at the business component layer receives and utilizes the aforementioned primitive attribute library to support user parameterization modifications. This component integrates a process rule library containing national standards, industry specifications, and enterprise-customized rules, and is configured with a rule self-evolution mechanism based on reinforcement learning agents. This mechanism dynamically optimizes conflict verification rules by defining state space, action space, and reward function. Simultaneously, the platform can connect to the factory's digital twin system, collecting field data via the OPC UA protocol to achieve a closed-loop verification between design parameters and actual field conditions.

[0051] Then, in the dynamic simulation phase of the process flow, the dynamic simulation analysis component of the business component layer calls the process simulation engine of the intelligent engine layer. This engine takes the edited and verified P&ID data as input, uses the sequential modular method to perform steady-state simulation to solve the material and energy balance, and uses the finite difference method to perform dynamic simulation. By establishing unit dynamic differential equations and discretizing them for solution, it simulates parameter changes under dynamic operating conditions such as feed flow fluctuations and sudden temperature rises, providing quantitative data support for risk analysis.

[0052] Next, HAZOP risk intelligent reasoning is performed. The HAZOP risk reasoning component in the business component layer calls the HAZOP knowledge base model in the intelligent engine layer. This model is built on the Transformer architecture and integrates risk cases and a guiding vocabulary that have undergone data cleaning and feature extraction. It extracts abnormal parameters from the process simulation results, matches them with the HAZOP deviation library, performs attention matching through the encoder, outputs possible causes, predicted consequences, and correlates rectification suggestions, generating a complete "deviation-cause-effect-suggestion" reasoning chain.

[0053] Finally, in the automatic generation stage of the structured report, the report generation component of the business component layer integrates risk reasoning results, process simulation parameters and basic information of P&ID drawings, and calls the natural language generation (NLG) module of the intelligent engine layer. Combined with the visualization engine, it automatically generates a structured report containing project overview, risk heat map, causal chain list and parameter evaluation analysis, and outputs it in PDF, Word or HTML format through the human-computer interaction layer.

[0054] Through the collaborative work based on the platform architecture described above, this invention achieves end-to-end automation from drawing parsing to report generation. To illustrate each technical detail in the above process in more detail, the specific implementation of this invention will be presented below:

[0055] Step 1: Intelligent analysis of chemical engineering drawings (breaking through the bottleneck of "low-quality recognition + implicit topology")

[0056] Execution Entity: AI-P&ID Recognition Engine (Software Program, belonging to the Intelligent Engine Layer)

[0057] operate:

[0058] 1. Core Model Architecture: Based on the improved YOLO v8 algorithm, the Diffusion-YOLO joint model is constructed by integrating the Diffusion generative model to achieve integrated low-quality drawing restoration and primitive detection. The Diffusion model adopts the UNet-2D architecture (containing 5 layers each for the encoder and decoder, with a 3×3 convolutional kernel size), while the YOLOv8 basic architecture is CSPDarknet53 (containing a backbone, neck, and detection head three-layer structure). The two share the 4th layer of the Diffusion model encoder and the C3-4 layers of the YOLOv8 backbone. The feature dimension is unified through 1×1 convolutional kernel mapping, which realizes parameter reuse and reduces redundant calculations.

[0059] The YOLOv8 improved algorithm specifically addresses three aspects of the improvements:

[0060] (1) Attention mechanism embedding: Add the CBAM (Convolutional Block Attention Module) attention module after the C3 module of the YOLOv8 backbone. First, strengthen the feature weights of key primitives (such as safety valves and pressure gauges) through the channel attention branch (composed of global average pooling + fully connected layers), and then locate primitive regions through the spatial attention branch (composed of convolutional layers + sigmoid activation function), reducing background noise interference and improving the accuracy of small target detection by 12%.

[0061] (2) Feature fusion optimization: The traditional top-down / bottom-up feature fusion path of PAN-FPN is changed to bidirectional cross fusion (BiFPN). High-resolution features (small primitives) and low-resolution features (large primitives) are weighted and fused to solve the problem of missed detection caused by the difference in symbol scale between pipelines and equipment, and the overall detection recall rate is improved by 8%.

[0062] (3) Improved loss function: The original IoU loss is replaced by CIoU (Complete Intersection over Union) loss function. The distance between the center point of the predicted box and the true box and the aspect ratio factor are introduced to make the bounding box regression more accurate and reduce the localization error by 15%.

[0063] 2. Low-quality drawing repair: For faded or creased drawings, first use Diffusion to generate a model and perform a "denoising-generation" process to repair blurred symbols (such as broken pipes) and complete faded annotations (such as instrument range text); the repaired drawings are then input into an improved YOLOv8 model for primitive detection to improve the recognition effect of low-quality drawings.

[0064] The specific steps of the "denoising-generation" process, based on the iterative repair process of the Diffusion model, are as follows:

[0065] (1) Noise injection: Gaussian noise (noise intensity decreases with each iteration) is randomly injected into the input low-quality drawings (including faded and creased images) to simulate the degradation process of "clear image → noisy image".

[0066] (2) Noise prediction: The encoder of the UNet-2D network extracts the features of the noisy image, and the decoder predicts the current noise distribution. In each iteration, the noise is predicted and subtracted (the noise reduction intensity is gradually increased).

[0067] (3) Detail repair: Customized repair strategies for chemical drawings - for broken pipes (pixel continuity interruption), the lines are completed by edge connection algorithm; for faded annotation text (pixel gray value is below the threshold), clear text area is generated by combining OCR recognition results; for cross creases (straight line noise), Hough transform is used to detect and eliminate them.

[0068] (4) Converging output: After 50 iterations, output a clear drawing after noise reduction, with a crease elimination rate of ≥90% and a faded text recognition accuracy of ≥85%.

[0069] 3. Image preprocessing: The following steps are performed sequentially: grayscale conversion (using weighted average method: R×0.299+G×0.587+B×0.114, converting to 8-bit grayscale image to eliminate color interference), denoising (bilateral filtering [spatial sigma=15, grayscale sigma=30], preserving edges while removing ink spots and crease noise), edge enhancement (Canny operator to extract contours), and block cutting (slicing into fixed-size sub-images to adapt to model input).

[0070] 4. Primitive Detection and Parameter Extraction: The YOLOv8 model is improved by embedding the CBAM attention module (including channel attention and spatial attention) to enhance the distinguishability of primitive features; based on transfer learning training, the pre-training set is an industrial P&ID annotation map (covering 200 categories of primitives in the GB / T24742 standard), and the historical drawings of the target enterprise are incorporated during the fine-tuning stage to adapt to the enterprise's symbol habits; the coordinate frame, category label and confidence of primitives such as equipment, pipelines and instruments are detected and output, and the text of primitive annotations is recognized by OCR at the same time to extract parameters such as equipment tag number, pipeline specification and instrument range, and stored in the primitive attribute library.

[0071] 5. Topology Construction: Using equipment, pipelines, and instruments as core nodes, a graph knowledge graph is constructed (incorporating the graph connection rules of GB / T24742 standard); a graph neural network (GNN) (2-layer GAT network) is used to learn the implicit design logic and automatically complete the unlabeled topology (such as spare branches); the connection relationship between pipelines and equipment is determined by coordinate deviation, and the medium flow direction is deduced by combining arrow symbols and pipeline numbering rules to construct the "equipment-pipeline-instrument" association matrix.

[0072] Function: To automate the parsing of chemical engineering drawings, reduce reliance on manual drawing processing, extract element parameters and topological relationships, and provide a data foundation for subsequent design and analysis.

[0073] Step 2: P&ID parameterized editing and conflict verification (achieving "rule evolution + virtual-real closed loop")

[0074] Execution Entity: P&ID Editing Module (Software Program, Belonging to the Business Component Layer)

[0075] operate:

[0076] 1. Data Import and Editing: Import the element attribute library data generated in step 1 with one click, support parametric modification (such as adjusting pipe diameter and equipment model), and retain modification records for easy traceability.

[0077] 2. Rule base integration: Integrate process rule base (including three types of data: national standards, industry specifications, and enterprise-customized rules, such as GB / T 24742-2009, API 521, etc.) to provide rule basis for conflict verification.

[0078] 3. Rule Self-Evolution: Based on reinforcement learning agents, rules are dynamically optimized. A state space (including the current rule base version, historical false negative rate, and historical false positive rate) is defined, along with an action space (adjustment of rule verification thresholds and ranking of rule priorities) and a reward function (comprehensively considering false negative rate, false positive rate, and design cycle compression rate to guide the agent in optimizing rules). The agent is updated after verifying a fixed number of drawings, and the Q-learning algorithm iteratively optimizes the rules to improve the accuracy of conflict detection.

[0079] 4. Virtual-to-real verification closed loop: Connect to the factory's digital twin platform and collect on-site data in real time (such as actual installation spacing and equipment vibration values) through the OPCUA protocol to verify the compatibility between design parameters and on-site working conditions, and discover hidden conflicts that traditional methods may miss in advance (such as design spacing being compliant but on-site pipelines being too dense to install).

[0080] Function: Enables parametric editing and automated conflict checking of P&ID drawings, dynamically optimizes check rules, and ensures the compliance and on-site adaptability of design schemes.

[0081] Step 3: Dynamic simulation of the process flow

[0082] Execution Entity: Process Simulation Module (Software Program, belonging to the Business Component Layer)

[0083] operate:

[0084] 1. Data Input: Using the P&ID data edited in step 2 as the input source, extract core parameters such as equipment model, media composition, and operating pressure / temperature.

[0085] 2. Simulation Algorithm: Steady-state simulation adopts the Sequential Modular Method (SM), which calculates material balance (mass conservation) and energy balance (heat conservation) based on the equation of state; dynamic simulation adopts the Finite Difference Method (FDM), which divides the equipment into unit models such as reactors, heat exchangers, and storage tanks, and establishes dynamic differential equations to simulate parameter changes under dynamic operating conditions.

[0086] 3. Simulation calculation: First, perform steady-state calculation and iteratively solve the equilibrium equation until convergence; then perform dynamic disturbance simulation to simulate typical dynamic working conditions (such as feed flow fluctuations and sudden temperature rises) and output parameter change curves.

[0087] Function: To quantify process parameters, provide dynamic operating condition data support for subsequent risk analysis, and identify abnormal trends in process parameters in advance.

[0088] The concepts and algorithms involved in steady-state simulation, dynamic simulation, element model, and state equations are as follows:

[0089] (1) Steady-state simulation: refers to the simulation of the equilibrium state of each parameter (temperature, pressure, flow rate, etc.) of the process system without changing with time, which is used to verify the material and energy balance under the design conditions.

[0090] Key algorithms and calculation steps:

[0091] Sequential Module Method (SM) – Steady-State Simulation:

[0092] Step 1: Arrange the unit models according to the process flow sequence (e.g., raw materials → pumps → heat exchangers → reactors → products) and define the material flow variables (flow rate, composition, temperature, etc.).

[0093] Step 2: Initialize the parameters of each unit (such as heat exchanger heat transfer area and reactor volume), assuming the initial material parameters.

[0094] Step 3: Calculate the material balance (∑Fin,i=∑Fout,i, where F is the component flow rate) and energy balance (∑Qin+∑Hin=∑Qout+∑Hout, where Q is heat and H is enthalpy) for each unit in sequence.

[0095] (2) Dynamic simulation: refers to the simulation of non-equilibrium state of system parameters changing over time, used to analyze the impact of disturbances (such as feed fluctuations and equipment failures) on the system and predict the trend of parameter changes.

[0096] Key algorithms and calculation steps:

[0097] Step 1: Discretize the time domain into small time steps (Δt=0.1s), and establish the element dynamic differential equation (e.g., the tank level equation). (V is the liquid level height).

[0098] Step 2: Discretize the differential equation using the Euler method (V(t+Δt)=V(t)+(Fin-Fout)·Δt).

[0099] Step 3: Apply disturbance conditions (such as a sudden increase of 20% in Fin), calculate parameter values ​​at each time step, and generate dynamic curves (such as the liquid level change curve over time).

[0100] Step 4: Set a safety threshold (e.g., 80% of the upper limit of the liquid level). When the simulation parameters exceed the threshold, trigger an alarm and calculate the alarm response time.

[0101] (3) Unit model: The chemical plant is broken down into independent functional units (such as reactors, pumps, heat exchangers, etc.). Each unit is based on the physical and chemical principles to establish a mathematical model that reflects the relationship between input and output parameters.

[0102] (4) Equation of state: Mathematical equations describing the state (such as density and enthalpy) of a substance under different temperatures and pressures. This platform adopts the PR-BM (Peng-Robinson-Benedict-Webb-Rubin) equation, which is suitable for phase equilibrium calculations of multi-component hydrocarbon mixtures.

[0103] Step 4: HAZOP Risk Intelligent Reasoning

[0104] Execution Entity: HAZOP Knowledge Base Model (software program, belonging to the Intelligent Engine Layer; HAZOP: Hazard and Operability Study)

[0105] operate:

[0106] 1. Model Architecture: Based on the Transformer architecture, it contains 6 layers of encoder (each layer contains a multi-head attention layer and a feed-forward layer) and 3 layers of decoder; the number of multi-head attention heads is 8, and the "causal mask + padding mask" strategy is adopted to block invalid data interference; the activation function is GELU, and a dropout layer is set to reduce overfitting.

[0107] 2. Knowledge Base Construction: Integrate risk cases and guiding terminology, perform data cleaning (removing duplicate and erroneous cases) and feature extraction (combining text features and process simulation parameters) on risk cases to form a standardized knowledge base to support risk reasoning.

[0108] 3. Risk Reasoning Process: Extract abnormal parameters from the simulation results in step 3 and match them with the HAZOP deviation library (such as "overheating" and "sudden pressure rise"); perform attention matching between deviation features and historical case features through the Transformer encoder, and output possible causes and confidence levels; predict the severity of consequences based on the cause-related equipment failure model; call the guiding dictionary to output targeted rectification suggestions and generate a complete chain of "deviation-cause-effect-suggestion".

[0109] Function: To automate the reasoning of chemical risks, reuse expert experience, and reduce the omissions and subjective biases of manual risk analysis.

[0110] Step 5: Automatic generation of structured reports

[0111] Execution Entity: Automatic Report Generation Module (Software Program, belonging to the Business Component Layer)

[0112] operate:

[0113] 1. Data Integration: Extract risk reasoning results (including deviations, causes, consequences, and recommendations) from step 4, extract process simulation parameters from step 3, and extract basic information of P&ID drawings from step 1, and integrate them into a report data source.

[0114] 2. Report Generation: Combining Natural Language Generation (NLG) technology with a visualization engine, the system generates structured reports. The reports include modules such as project overview, risk heatmap (risk levels are marked based on P&ID drawings), causal chain list, parameter evaluation and analysis (comparison of simulated parameters and design values), and rectification plan. The system supports output in PDF, Word, HTML and other formats, and can be adapted to customized templates for enterprises.

[0115] 3. Visual presentation: The risk heat map uses black and white lines (high-density diagonal lines indicate high risk, medium-density diagonal lines indicate medium risk, and low-density diagonal lines indicate low risk). The parameter comparison curves are distinguished by solid lines (simulated values) and dashed lines (design values), which meets the requirements of the patent drawings.

[0116] Function: Automatically generates standardized risk analysis reports, reducing the workload of manual report writing and improving the standardization and readability of reports.

[0117] The various embodiments described in this specification are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An integrated platform device for chemical engineering design and safety risk analysis, characterized in that, The platform includes: The data foundation layer is used to integrate multi-source data and perform data cleaning and structure transformation; The intelligent engine layer, which communicates with the data base layer, is used to execute core algorithm processing. It includes an AI-P&ID recognition engine, a process simulation engine, and a large model of the HAZOP knowledge base. The AI-P&ID recognition engine realizes primitive detection and low-quality drawing repair of chemical drawings based on a joint model built on an improved YOLO algorithm and a Diffusion generation model. The improved YOLO algorithm embeds a CBAM attention module, adopts a bidirectional cross-fusion feature fusion path, and a CIoU loss function. The business component layer, which communicates with the intelligent engine layer, is used to call the core algorithm to process the results and realize core business functions. It includes a P&ID intelligent parsing component, a P&ID editing and verification component, a dynamic simulation analysis component, a HAZOP risk reasoning component, and a report generation component. The P&ID editing and verification component integrates a process rule library and is configured with a rule self-evolution mechanism based on reinforcement learning for dynamically optimizing conflict verification rules. The rule self-evolution mechanism is based on reinforcement learning agents to dynamically optimize rules. It defines a state space, including the current rule base version, historical false negative rate, and historical false positive rate; an action space, including rule verification threshold adjustment and rule priority ranking; and a reward function, which comprehensively considers the false negative rate, false positive rate, and design cycle compression rate to guide the agent in optimizing rules. The agent updates after verifying a fixed number of drawings, iteratively optimizing rules through a Q-learning algorithm to improve conflict detection accuracy. It also includes a virtual-real verification closed loop: connecting to the factory's digital twin platform, collecting on-site data in real time through the OPCUA protocol, including actual installation spacing and equipment vibration values, verifying the compatibility of design parameters with on-site working conditions, and discovering hidden conflicts that traditional methods may miss in advance; The human-computer interaction layer communicates with the business component layer and is used to provide a user interface and support interface with external systems.

2. The integrated platform device for chemical engineering design and safety risk analysis according to claim 1, characterized in that, The data base layer uses distributed database technology to achieve data storage and governance.

3. The integrated platform device for chemical engineering design and safety risk analysis according to claim 1, characterized in that, The process simulation engine is configured to perform steady-state simulations based on the sequential module method and dynamic simulations based on the finite difference method.

4. The integrated platform device for chemical engineering design and safety risk analysis according to claim 1, characterized in that, The HAZOP knowledge base model is built on the Transformer architecture and integrates risk cases and a guide vocabulary that have undergone data cleaning and feature extraction.

5. The integrated platform device for chemical engineering design and safety risk analysis according to claim 1, characterized in that, The report generation component is configured to integrate risk reasoning results, process simulation parameters, and basic information from P&ID drawings, and combine natural language generation technology with a visualization engine to output a structured report.

6. The integrated platform device for chemical engineering design and safety risk analysis according to claim 1, characterized in that, The platform is configured to perform the following automated processes: The AI-P&ID recognition engine intelligently analyzes chemical engineering drawings, extracts primitive parameters, and constructs topological relationships. The P&ID editing and verification component performs parametric editing and conflict verification on the parsed drawing data; The process simulation engine performs dynamic simulation of the process flow based on the verified data; The HAZOP risk reasoning component performs intelligent risk reasoning based on simulation results; The report generation component automatically generates a structured risk analysis report.

7. The integrated platform device for chemical engineering design and safety risk analysis according to claim 6, characterized in that, The AI-P&ID recognition engine performs intelligent parsing of chemical engineering drawings, including: using a joint model based on an improved YOLO algorithm and a Diffusion generation model to perform integrated processing of low-quality drawing repair and primitive detection; the P&ID editing and verification component performs conflict verification, including a self-evolutionary process based on reinforcement learning agents to achieve dynamic rule optimization.

Images