A papermaking equipment fault tracing method and system based on a process knowledge graph

Abstract

The application relates to the technical field of intelligent manufacturing, and discloses a papermaking equipment fault tracing method and system based on a process knowledge graph, the method and system comprising data acquisition and preprocessing, papermaking process knowledge graph construction, fault event detection and matching, fault tracing and propagation path reasoning, and maintenance scheme recommendation and optimization. The application overcomes the limitations of traditional methods in capturing deep-level cause-and-effect correlations and fault propagation paths across equipment, processes and time sequences in papermaking equipment. By constructing a comprehensive papermaking process knowledge graph, heterogeneous knowledge such as equipment operation state data, process parameter data, production quality data, equipment structure principles, process flow knowledge, fault mode knowledge and maintenance experience is deeply fused and semantically associated, so that the system can go beyond the surface correlation of data and deeply mine the internal mechanism and propagation chain of faults.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and discloses a method and system for tracing the source of faults in papermaking equipment based on process knowledge graphs. Background Technology

[0002] The diagnosis and tracing of papermaking equipment malfunctions primarily relied on traditional experience-based judgment, regular maintenance, and alarm systems based on simple threshold settings. With the development of industrial automation and information technology, many advanced technologies have been introduced into the field of fault diagnosis. Early practices mainly included manual inspections and regular maintenance. Maintenance personnel, relying on their rich practical experience and professional knowledge, used sensory means such as listening, seeing, touching, and smelling to make preliminary judgments about the equipment status, and combined this with the equipment manual and historical maintenance records to troubleshoot. Building on this foundation, methods based on sensor data acquisition and analysis have gradually developed. By monitoring key operating parameters such as vibration, temperature, pressure, and current, and combining statistical process control methods or simple expert rule systems, abnormal parameters or deviations from preset thresholds can be identified, thereby triggering early warnings or performing preliminary fault location.

[0003] Furthermore, with the rise of big data and machine learning technologies, some research and applications have begun to utilize historical fault data to train predictive models, aiming to make predictions before faults occur or to assist in diagnosis through pattern matching after faults occur. These methods have improved the efficiency and accuracy of fault diagnosis to some extent, especially in dealing with single equipment anomalies with clear characteristics or known fault patterns, demonstrating their inherent value and contribution.

[0004] However, as paper production lines become increasingly complex and intelligent, and more stringent requirements are placed on production continuity and efficiency, some inherent characteristics of the aforementioned technical solutions at the principle level are gradually revealing their limitations in addressing new challenges, and their deep-seated contradictions are becoming increasingly prominent. The reason for this is that failures in papermaking equipment are not isolated events, but often involve complex interactions and chain reactions between multiple pieces of equipment and multiple process parameters.

[0005] Traditional diagnostic methods, whether based on single-parameter thresholds, local expert rules, or pattern recognition models trained on historical data, often struggle to effectively capture and understand deep-seated causal relationships and fault propagation paths across equipment, processes, and timelines. While statistical or machine learning models can identify correlations from massive amounts of data, they often fail to reveal the underlying semantic logic and intrinsic mechanisms. A seemingly simple pump vibration anomaly may not be rooted in the pump itself, but rather in upstream pipeline blockage causing a sudden pressure surge, downstream valve incomplete opening leading to back pressure, or even batch-to-batch variations in raw materials causing changes in slurry viscosity that affect pump load. Such complex causal chains across systems and domains cannot be accurately inferred and traced based solely on correlations between data without a clear knowledge framework.

[0006] Furthermore, the highly specialized and tacit nature of papermaking process knowledge presents significant challenges to the construction of traditional expert systems. Expert experience often exists in unstructured forms in maintenance manuals, operating procedures, fault logs, and the minds of senior engineers, making it difficult to systematically and standardizedly integrate and utilize this knowledge. Simply building a rule base is not only time-consuming and labor-intensive but also struggles to cover all possible fault modes and anomalies, especially when facing new equipment, new processes, or complex faults, where its diagnostic capabilities significantly decline. More critically, existing technologies generally lack a mechanism capable of deeply integrating, semantically associating, and logically reasoning with multi-source heterogeneous knowledge, including real-time equipment operating status data, historical fault cases, equipment structural principles, process flow knowledge, and maintenance experience.

[0007] Without such a mechanism, even with a massive IoT data warehouse, this data is like scattered pearls, unable to be automatically strung together into a complete chain reflecting the occurrence, development, and propagation of faults. This results in a lengthy fault tracing process, diagnostic accuracy limited by experience, and even repeated occurrences of the same fault that are difficult to eradicate. This situation of data silos and knowledge gaps severely restricts the level of intelligence and automation in fault tracing of papermaking equipment, forcing maintenance personnel to spend a lot of time on manual troubleshooting and trial and error when facing complex faults, thus significantly increasing downtime and maintenance costs.

[0008] Therefore, how to construct a technical solution that can effectively integrate multi-source heterogeneous data and domain expert knowledge, and can achieve rapid and accurate tracing of the root causes of papermaking equipment failures and clearly present their propagation paths based on deep semantic association and logical reasoning, has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0009] To achieve the above-mentioned objectives, this invention provides a method and system for tracing the root causes of papermaking equipment failures based on process knowledge graphs. It aims to solve the technical challenges in the existing technology of effectively capturing and understanding deep causal relationships and failure propagation paths across equipment, processes, and time series during the fault tracing process of papermaking equipment, as well as the difficulty in deeply integrating multi-source heterogeneous data with domain expert knowledge, semantic association, and logical reasoning. This enables rapid and accurate tracing of the root causes of papermaking equipment failures and clearly presents their propagation paths.

[0010] The present invention proposes a fault tracing method for papermaking equipment based on process knowledge graphs, which includes the following steps: S1, Data Acquisition and Preprocessing: Construct a multi-source data acquisition network to collect real-time operating status data, process parameter data, production quality data, environmental parameter data, and historical fault and maintenance data of each core and auxiliary equipment during the papermaking process; perform cleaning, standardization, missing value filling, outlier handling, and time series alignment preprocessing operations on the collected raw data, and store the processed data in an integrated data warehouse. S2, Construction of Papermaking Process Knowledge Graph: Extract entities and relationships from structured and unstructured data sources, construct an ontology of the papermaking process domain, and then construct a papermaking process knowledge graph; the papermaking process knowledge graph uses nodes to represent entities in the papermaking production process and edges to represent semantic relationships between entities; S3, Fault Event Detection and Matching: Continuously monitor the real-time operating data in the integrated data warehouse, and detect equipment anomalies and fault events through a preset fault diagnosis model; when a fault event is detected, extract its key features and match them with known fault modes in the papermaking process knowledge graph; S4, Fault source tracing and propagation path reasoning: Starting from the matched fault type or fault phenomenon node, perform association query and logical reasoning in the papermaking process knowledge graph, combine real-time data and historical case data, traverse the graph along the semantic relationship edge, gradually trace to the potential root cause of the fault, and construct the fault propagation path. S5, Maintenance Solution Recommendation and Optimization: Based on the identified root causes and propagation paths of the faults, a correlation query is performed in the papermaking process knowledge graph to extract maintenance experience, maintenance operating procedures, required tools and spare parts information related to the fault type, equipment model, root cause, and historical maintenance records, and to automatically generate structured maintenance solution recommendations; the method also includes: further providing a user feedback interface, and continuously optimizing and iterating the relation weights, inference rules, and maintenance solution recommendation model in the papermaking process knowledge graph based on the actual maintenance results of the maintenance personnel.

[0011] Preferably, the multi-source data acquisition network is implemented by deploying various industrial sensors and data acquisition terminals; The sensors are interconnected with the edge computing gateway and host computer system via industrial Ethernet, wireless sensor network or fieldbus to achieve high-frequency, low-latency data transmission; The collected operating status data includes vibration signals, temperature, pressure, flow rate, current, voltage, rotational speed, and noise intensity. The process parameter data includes pulp concentration, pH value, sizing degree, fiber length, filler content, white water turbidity, wire section vacuum degree, press line pressure, drying cylinder surface temperature, drying section steam pressure, and paper web tension. The production quality data covers paper basis weight, thickness, moisture content, whiteness, tensile strength, tear resistance, and uniformity. The historical fault and maintenance data includes fault codes, fault description text, occurrence time, duration, maintenance operation records, information on replaced parts, and feedback from maintenance personnel.

[0012] Preferably, the preprocessing operations of cleaning, standardizing, filling missing values, handling outliers, and aligning time series data for the collected raw data include, but are not limited to: the data cleaning uses median filtering based on sliding window, wavelet denoising technology, and clustering-based outlier detection algorithm to identify and correct outliers; The missing value imputation employs interpolation or regression prediction methods based on the data type and missing value pattern; the time series alignment achieves synchronization of data at different frequencies through precise timestamp matching and resampling techniques. The integrated data warehouse is built on a distributed storage architecture, uses columnar storage database technology for efficient reading and writing of time-series data, uses a relational database to store structured metadata, and integrates a document database to store unstructured logs.

[0013] Preferably, the knowledge extraction process in step S2 is implemented using natural language processing technology and deep learning models, including named entity recognition, relation extraction and event extraction; The named entity recognition module is based on a pre-trained language model of the Transformer architecture, and is fine-tuned by combining a conditional random field layer or a pointer network to identify equipment components, fault types, fault phenomena, fault causes, process parameters, maintenance measures, tools and consumables entities in the text. The relationship extraction module uses graph neural networks or deep learning models based on attention mechanisms to learn the association patterns between entities. The papermaking process ontology defines all core concepts, attributes, and their interrelationships in the papermaking production process, providing a unified semantic framework; the papermaking process knowledge graph is stored in a graph database and uses an attribute graph model.

[0014] Preferably, the fault diagnosis model includes, but is not limited to: a control chart method based on statistical process control, used to monitor the mean and volatility of key process parameters; Machine learning-based anomaly detection algorithms are used to identify nonlinear anomaly patterns in multivariate time series data; rule-based expert systems are used to identify complex anomalies that meet specific logical conditions. The matching process is achieved by calculating the similarity between the fault event feature vector and the standardized fault pattern entities stored in the knowledge graph, using a vector similarity matching method based on semantic embedding or a graph query algorithm based on attribute and relation pattern matching.

[0015] Preferably, the association query and logical reasoning process includes: Initial root cause exploration: Starting from the initial fault phenomenon node, the knowledge graph is explored upstream using a graph traversal algorithm to find the direct cause node; at the same time, the exploration path is pruned or weighted based on the abnormal correlation of parameters in the real-time data. Multi-level causal chain construction: For the identified direct cause nodes, continue to perform recursive tracing to delve deeper into the root causes; in this process, the weights and priorities of different types of relationship edges are comprehensively considered; Historical case verification and supplementation: The potential failure paths and root causes formed during the reasoning process are compared and verified with historical failure cases in the integrated data warehouse; Multi-source information fusion reasoning: In the reasoning process, the real-time sensor data and the structured knowledge in the knowledge graph are dynamically fused, and the real-time data is used as a constraint or verification evidence for reasoning. Root cause priority assessment: For all identified potential root causes, a comprehensive assessment is conducted based on their centrality in the knowledge graph, their matching degree with the historical failure cases, the confidence level of the real-time data verification, and the severity of their impact on production, and priorities are assigned to determine the most likely set of root causes. Fault propagation path construction and visualization: Based on the reasoning results, a complete fault propagation path from the fault phenomenon to the final root cause is constructed, and the propagation path is presented in a graphical manner on the user interface.

[0016] This invention discloses a fault tracing system for papermaking equipment based on process knowledge graphs, comprising: The data acquisition module is used to configure multi-channel data acquisition units and various industrial sensor arrays to collect various operating data on the paper production line in real time. The data preprocessing and storage module is used to receive the raw data transmitted by the data acquisition module, and to perform data cleaning, standardization, data fusion and storage on it; The knowledge graph construction and management module is used for the creation, updating, and maintenance of the papermaking process knowledge graph; The fault event detection module is used to analyze the data in the data preprocessing and storage module in real time to identify abnormal equipment operation or fault events. The fault tracing and reasoning module is used to perform in-depth fault causal reasoning using the papermaking process knowledge graph. The maintenance decision support and visualization module is used to present the fault tracing results to the user in an intuitive way and provide intelligent maintenance solution recommendations. The system feedback and optimization module is used for continuous learning and optimization of the entire system.

[0017] Preferably, the data acquisition module includes: the industrial sensor array including but not limited to piezoelectric accelerometers, resistance temperature sensors, capacitive pressure sensors, electromagnetic flowmeters, clamp-on current transformers, photoelectric speed sensors, MEMS microphones, and industrial vision cameras installed in key parts of each papermaking equipment. The data acquisition unit integrates an analog-to-digital converter, a signal conditioning circuit, and a data buffer, and transmits data with the data preprocessing and storage module through an industrial Ethernet interface or a wireless communication module. The data acquisition module supports multiple communication protocols. The data preprocessing and storage module includes: the data cleaning function includes median filtering based on sliding window, wavelet denoising, and outlier detection algorithm based on machine learning; the data standardization function includes Min-Max standardization or Z-score standardization. The data fusion function integrates data from different data sources through timestamp alignment algorithms and semantic association algorithms based on entity IDs; the storage unit includes high-performance time-series databases, relational databases, and unstructured data storage.

[0018] Preferably, the knowledge graph construction and management module includes: The knowledge extraction unit is used to perform named entity recognition using a pre-trained deep learning model to identify specific entity types in the papermaking field, and to extract relationships using a model based on graph neural networks or multi-head attention mechanisms to identify semantic relationships between entities; the structured data parser maps structured information into nodes and edges of the knowledge graph. The ontology construction unit is used to provide domain ontology modeling tools, define the core concepts, attributes and relational patterns of the papermaking production domain, and ensure the semantic consistency and scalability of the knowledge graph. The knowledge fusion unit is used to integrate knowledge from different sources and formats to resolve entity disambiguation and relationship conflict issues. The knowledge storage and query unit is used to store the constructed knowledge graph using a high-performance graph database and to provide an efficient graph query language interface. The knowledge update and maintenance unit provides a user interface for manual editing and proofreading of knowledge, and has an automatic or semi-automatic knowledge incremental update mechanism.

[0019] Preferably, the fault tracing and reasoning module includes: The starting node matching unit is used to receive the output of the fault event detection module and, based on the key information in the fault event report, match the most relevant fault phenomenon node or fault type node in the knowledge graph as the starting point for tracing the source. The graph traversal and path discovery unit is used to execute a customized graph traversal algorithm in the knowledge graph, starting from the matched starting node. The algorithm combines the characteristics of breadth-first search and depth-first search, prioritizing the exploration along edges with strong causal relationships that cause, trigger, or influence. It also considers path length, node importance, and edge weight to optimize the search direction and dynamically prunes paths that do not conform to real-time data logic or are unreasonable. A logic reasoning engine is used to embed a rule-based inference engine and a graph-based reasoning algorithm. The rule-based inference engine performs deductive reasoning based on predefined Datalog or Prolog rules, and the graph-based reasoning algorithm uses node and edge representations learned by graph neural networks or knowledge graph embedding models to perform link prediction or node classification. The multi-source information fusion reasoning unit is used to dynamically acquire real-time running data and historical case data related to the current reasoning path during the reasoning process, and use the real-time data as constraints or verification evidence for reasoning. The root cause identification and priority assessment unit is used to comprehensively evaluate all potential root causes obtained through reasoning, and calculate their confidence or priority. The assessment indicators include the reliability of the tracing path, the degree of consistency with real-time data, the corroboration strength of historical cases, the severity of the impact on production, and the maintenance cost. The fault propagation path construction unit is used to construct a complete directed graph structure from the fault phenomenon to the most likely root cause based on the results of the root cause identification and priority evaluation.

[0020] The papermaking equipment fault tracing method and system based on process knowledge graph proposed in this invention can achieve the following beneficial effects through the above technical solution: 1. This invention overcomes the limitations of traditional methods in capturing deep-seated causal relationships and fault propagation paths across equipment, processes, and timelines in papermaking equipment. By constructing a comprehensive papermaking process knowledge graph, it deeply integrates and semantically correlates heterogeneous knowledge such as equipment operating status data, process parameter data, production quality data, equipment structural principles, process flow knowledge, fault mode knowledge, and maintenance experience. This allows the system to go beyond the surface correlation of data and delve into the intrinsic mechanisms and propagation chains of faults.

[0021] 2. This invention significantly improves the accuracy and efficiency of fault tracing in papermaking equipment. By combining real-time data with structured knowledge graphs and employing advanced algorithms such as graph traversal, pattern matching, and logical reasoning, it can quickly pinpoint single or complex deep-seated root causes from complex fault phenomena and clearly present their propagation paths. This greatly reduces the time spent on manual troubleshooting and trial and error, lowers the reliance on expert experience in fault diagnosis, and enables non-professionals to perform efficient fault tracing with the system's assistance.

[0022] 3. This invention makes papermaking process knowledge explicit, systematic, and reusable. Through natural language processing and deep learning technologies, tacit knowledge scattered across various documents and expert minds is extracted, structured, and integrated into a knowledge graph, forming a unified, machine-understandable knowledge system. This not only facilitates knowledge storage, retrieval, and sharing but also lays a solid foundation for subsequent intelligent applications.

[0023] 4. This invention provides an intelligent maintenance decision support and continuous optimization mechanism. Based on maintenance experience and expert suggestions in the knowledge graph, the system can automatically generate detailed and actionable maintenance plans, providing precise guidance for maintenance personnel. Simultaneously, through continuous user feedback and model optimization mechanisms, the system can continuously learn and iterate, enabling the knowledge graph and fault-finding model to improve over time and with the accumulation of new fault cases. This forms an adaptive and self-evolving intelligent diagnosis and maintenance platform, continuously improving the accuracy of fault tracing, reducing downtime and maintenance costs, and ensuring the continuous and stable operation of the paper production line. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the structure of a papermaking equipment fault tracing system based on process knowledge graph according to the present invention; Figure 2 This is a flowchart illustrating a method for tracing the source of papermaking equipment faults based on process knowledge graphs, as proposed in this invention. Figure 3 This is a schematic diagram of the structure of the papermaking process knowledge graph of the present invention; Figure 4 This is a schematic diagram illustrating the fault tracing and propagation path reasoning of the present invention; Figure 5 This is a schematic diagram of the data acquisition network of the present invention; Figure 6 This is a schematic diagram of the maintenance decision support and visualization interface of the present invention. Detailed Implementation

[0025] This invention discloses a method and system for tracing the root causes of papermaking equipment failures based on process knowledge graphs. It aims to address the increasing complexity of equipment in papermaking production sites, the challenges of fusion of multi-source heterogeneous data, and the high dependence of fault diagnosis on domain expert experience. This allows for rapid and accurate root cause tracing and clear propagation path presentation of papermaking equipment failures. This embodiment will elaborate on the specific implementation details, key technical principles, and effectiveness of this method and system in practical applications.

[0026] The fault tracing system for papermaking equipment proposed in this invention has the following overall architecture: Figure 1 As shown, the system includes a data acquisition module, a data preprocessing and storage module, a knowledge graph construction and management module, a fault event detection module, a fault tracing and reasoning module, a maintenance decision support and visualization module, and a system feedback and optimization module. These modules work closely together to form a closed-loop intelligent diagnosis and maintenance system.

[0027] Example 1: The data acquisition module is the perception layer of the entire system, responsible for acquiring various types of data involved in the papermaking process in real time, continuously, and accurately. This module is equipped with an array of various industrial sensors, which are strategically installed in key parts of the papermaking equipment. Piezoelectric accelerometers are deployed at locations such as the refiner motor bearings, press rolls, dryer cylinder surfaces, and doctor blade drive mechanisms to monitor the vibration spectrum characteristics of the equipment; resistance temperature sensors are used to acquire the temperature distribution on or inside the surface of components in real time; capacitive pressure sensors are used to monitor pressure fluctuations in the pulp pumping system or press unit; electromagnetic flowmeters are used to accurately measure the flow rate of pulp, white water, or steam; clamp-on current transformers are used to monitor the current signal of the motor load and operating status; photoelectric speed sensors are used to measure the rotational speed of the rolls or drive shafts; MEMS microphones can capture abnormal sound characteristics of the equipment; and industrial vision cameras can monitor paper web forming, breakage, or surface defects in real time.

[0028] These sensors digitize and perform preliminary data processing through a data acquisition unit integrating an analog-to-digital converter and signal conditioning circuitry. High-speed data transmission is then achieved with higher-level data processing units via an industrial Ethernet interface or wireless communication module. The data acquisition module is compatible with various mainstream industrial communication protocols, including but not limited to Modbus TCP / IP, OPC UA, MQTT, and EtherCAT, ensuring interoperability with papermaking equipment and automation systems from different manufacturers and generations. The acquired data features high frequency and low latency; vibration data can reach a sampling rate of 25.6 kHz, and temperature and pressure data can be updated in seconds, ensuring sensitive capture of dynamic changes in the equipment.

[0029] Example 2: The data preprocessing and storage module receives the raw data transmitted from the data acquisition module and performs a series of cleaning, standardization, fusion, and efficient storage operations on it. In the data cleaning stage, the system employs a multi-layered strategy. First, it effectively filters out high-frequency noise and random interference using median filtering and wavelet denoising algorithms based on sliding windows. Then, by deploying a machine learning-based outlier detection algorithm, it can identify and process outliers or abnormal readings caused by sensor malfunctions.

[0030] The data standardization function employs Min-Max or Z-score standardization methods to unify sensor data from different dimensions and ranges onto a comparable scale, eliminating the impact of dimensional differences on subsequent model training and analysis. The data fusion function is crucial; it synchronizes data from different sensors and devices along the time dimension using a precise timestamp alignment algorithm, and logically binds the scattered data to equipment entities and process steps in the papermaking process knowledge graph through an entity ID-based semantic association algorithm, thereby constructing a comprehensive and consistent multi-source data view.

[0031] In terms of data storage, this module constructs a hierarchical storage system: for massive amounts of real-time sensor data and historical operational data, the system adopts a high-performance time-series database to support high-concurrency writing, efficient time-range querying, and aggregation calculation; for structured metadata such as equipment asset information, process parameter configuration, and personnel organizational structure, a relational database is used for storage to ensure data transaction consistency and strong query capabilities; in addition, for unstructured data such as maintenance logs, operating procedure documents, and expert interview records, the system uses a distributed file system or document database for flexible storage to adapt to their diverse formats and content.

[0032] The knowledge graph construction and management module is responsible for the creation, updating, and continuous maintenance of the papermaking process knowledge graph, making the scattered and implicit knowledge in the papermaking field explicit and structured. This module consists of a knowledge extraction unit, an ontology construction unit, a knowledge fusion unit, a knowledge storage and query unit, and a knowledge update and maintenance unit. The knowledge extraction unit integrates an advanced text analysis engine and a structured data parser.

[0033] The text analysis engine utilizes pre-trained deep learning models, such as BERT-CRF or Bi-LSTM-CRF models based on the Transformer architecture, fine-tuned for papermaking terminology and context to achieve high-precision named entity recognition. It can accurately identify equipment components (such as pulp tank agitators, vacuum dewatering tanks, and calender rolls), fault types (such as pump idling, motor overheating, and paper web wrinkling), fault symptoms (such as increased vibration amplitude, reduced pulp output, and uneven paper thickness), potential causes (such as aging seals, insufficient bearing lubrication, and fiber entanglement), maintenance actions (such as impeller replacement, gap adjustment, and blockage clearing), process parameters (such as wire density, drying section humidity, and paper web tension), and required tools and spare parts from unstructured text. Simultaneously, the engine employs models based on graph neural networks or multi-head attention mechanisms for relation extraction, learning complex semantic relationships between entities.

[0034] The structured data parser is responsible for automatically converting structured information such as equipment parameter manuals, historical fault code tables, and SCADA system configuration data into nodes and edges in the knowledge graph through predefined mapping rules, ensuring the collaborative integration of knowledge from different sources. The ontology construction unit provides an intuitive set of domain ontology modeling tools, allowing domain experts to define core concepts, attributes, and their interrelationships in the papermaking process. This provides a unified semantic framework and classification system for the knowledge graph, ensuring logical consistency and scalability of knowledge. The knowledge fusion unit handles knowledge fragments from different data sources and extraction processes, ensuring the uniformity, lack of redundancy, and high accuracy of the knowledge graph through entity disambiguation and relation conflict resolution strategies.

[0035] The knowledge storage and query unit employs a high-performance graph database to store the constructed papermaking process knowledge graph. This type of database can efficiently manage and query complex graph-structured data and supports graph query languages ​​such as Cypher, Gremlin, or SPARQL, facilitating complex relational queries and multi-hop graph traversal operations. The knowledge update and maintenance unit provides a user-friendly interface for domain experts to manually edit and proofread knowledge, correcting extraction errors or supplementing newly discovered knowledge. More importantly, it features an automatic or semi-automatic incremental knowledge update mechanism, dynamically adjusting entities, relationships, and their attributes in the knowledge graph based on new fault cases, maintenance feedback, and changes in industry standards, thereby achieving continuous evolution and improvement of the knowledge graph.

[0036] The fault event detection module, as a crucial component of system early warning and initial diagnosis, continuously analyzes the massive amounts of data provided by the data preprocessing and storage module in real time to identify equipment malfunctions or potential fault events. This module integrates several advanced detection algorithms: First, a rule-based detector, using a pre-set set of expert experience rules, can quickly respond to known types of abnormal patterns. Second, a statistical analysis detector effectively detects statistical anomalies in data sequences using sliding window statistics monitoring, control chart methods, and time series decomposition.

[0037] Furthermore, the machine learning anomaly detector is one of the core capabilities of this module. It deploys supervised learning models to classify predefined fault modes, or unsupervised learning models to detect unknown or novel anomaly patterns. These models are trained offline using a large amount of historical normal operation data to learn the data patterns of equipment in a healthy state, and then make real-time predictions online, determining whether the current data is abnormal by reconstruction error, anomaly score, or classification probability. When an anomaly is detected, the feature extraction unit automatically extracts the contextual information of the fault occurrence, including the name of the abnormal parameter, the abnormal amplitude, the abnormal trend (such as continuous increase, periodic fluctuation), the associated equipment components, the exact time of the fault occurrence, and a preliminary description of the symptoms, and structures this information into a fault event report, providing accurate input for subsequent fault tracing.

[0038] The fault tracing and reasoning module utilizes a pre-constructed knowledge graph of papermaking processes for in-depth causal reasoning of faults. This module includes a starting node matching unit, a graph traversal and path discovery unit, a logical reasoning engine, a multi-source information fusion reasoning unit, a root cause identification and priority evaluation unit, and a fault propagation path construction unit. The starting node matching unit receives fault event reports from the fault event detection module. Based on the key information in the reports, it vectorizes the fault event features using semantic embedding technology and calculates similarity between these vectors and the embedding vectors of standardized fault phenomenon nodes or fault type nodes stored in the knowledge graph. This allows it to match the starting tracing node that is semantically most relevant to the current fault event.

[0039] The graph traversal and path discovery unit starts from the matched initial node and executes a customized graph traversal algorithm within the knowledge graph. This algorithm combines the advantages of breadth-first search and depth-first search. During the search process, the system prioritizes exploring edges with strong causal relationships such as cause, trigger, and influence, ensuring rapid identification of key causal chains. Simultaneously, the algorithm dynamically considers path length, node importance, and edge weights (based on historical failure frequency, expert experience confidence, or real-time data verification strength) to optimize the search direction and dynamically prunes paths that do not conform to real-time data logic or are clearly unreasonable. The logic reasoning engine embeds a rule-based inference engine and a graph-based reasoning algorithm. The rule-based inference engine performs deductive reasoning based on predefined Datalog or Prolog rule sets to derive new facts or verify potential causal chains.

[0040] Graph-based reasoning algorithms utilize node and edge representations learned from graph neural networks or knowledge graph embedding models to predict links, thereby discovering potential, unmodeled causal relationships, or classifying nodes to identify root cause types. During the reasoning process, the multi-source information fusion reasoning unit interacts in real-time with the data preprocessing and storage module, dynamically acquiring real-time operational data and historical case data related to the current reasoning path.

[0041] Real-time data is used as constraints or verification evidence for inference. When the knowledge graph infers that poor bearing lubrication is a potential cause, the system will immediately query the real-time temperature sensor data and vibration spectrum analysis results of the corresponding equipment. If the bearing temperature continues to rise and the vibration spectrum shows an increase in energy at a specific frequency, the reliability of the inference is further confirmed.

[0042] Conversely, if real-time data is inconsistent, the inference direction is adjusted or the priority of that path is reduced. Simultaneously, the system uses similar historical failure cases as references for inference, assisting in the discovery of potential, unmodeled causal chains and improving the comprehensiveness of the inference. The root cause identification and priority evaluation unit comprehensively evaluates all potential root causes obtained through inference, calculating their confidence level or priority. Evaluation indicators are multi-dimensional, including the reliability of the tracing path (weights on the path, centrality of nodes), consistency with real-time data, strength of corroboration from historical cases, severity of impact on production (downtime, output loss), and estimated maintenance costs.

[0043] Finally, one or more sorted root cause lists and their confidence levels are output for maintenance personnel to refer to. The fault propagation path construction unit constructs a complete directed graph structure from fault symptoms to the most likely root cause based on the results of root cause identification and priority assessment, clearly showing the chain of fault occurrence, development, and propagation. This path consists of a series of nodes (representing equipment, components, parameters, fault symptoms, and causes) and edges (representing different types of causal relationships and influence relationships), providing users with an intuitive panoramic view of the fault.

[0044] The maintenance decision support and visualization module is responsible for presenting fault tracing results to users in an intuitive and easy-to-understand manner, and providing intelligent maintenance solution recommendations. The visualization engine uses interactive graphical interface technology to dynamically display the fault propagation path in the form of a network graph. Nodes represent entities (such as equipment, sensors, fault phenomena, and root causes), and edges represent relationships (such as cause, impact, and composition). The color, size, and thickness of nodes and edges can be encoded according to their importance, type, or confidence level, and root cause nodes can be highlighted. This interface allows users to zoom, pan, and drag the graph, and click on nodes or edges to view detailed attribute information, such as equipment model, parameter range, and fault history.

[0045] Simultaneously, it provides historical trend charts of key parameters, real-time data dashboards, and views integrating 3D equipment models, helping users comprehensively understand the fault scenario from multiple dimensions. The maintenance solution recommendation engine, based on the root causes and propagation paths output by the fault tracing and reasoning module, automatically queries the knowledge graph for associated maintenance measure nodes, required tool nodes, spare parts information nodes, operating procedure nodes, and safety specification nodes, and generates structured and actionable maintenance solutions based on a preset recommendation algorithm.

[0046] The proposed solution is detailed, including thorough troubleshooting steps, necessary safety precautions, a recommended list of repair tools, required spare parts models and quantities, estimated repair time, and relevant expert advice. The alarm and notification unit integrates multiple alarm mechanisms, such as audible and visual alarms, SMS notifications, email notifications, or app push notifications, ensuring that fault information, troubleshooting results, and repair suggestions are promptly and accurately sent to relevant repair personnel or managers. The user feedback and performance evaluation interface provides a user-friendly interface, allowing repair personnel to input actual repair results, actual repair time, replaced parts, effectiveness evaluation, and any new knowledge discovered during the repair process. The data collected by this interface will serve as input to the system feedback and optimization module, forming a closed-loop feedback mechanism to continuously improve the system's intelligence.

[0047] The system feedback and optimization module is crucial for the system's continuous learning and adaptive evolution. The knowledge graph optimization unit incrementally updates, deletes, or corrects entities and relationships in the knowledge graph based on user feedback and historical maintenance results. If a causal chain is repeatedly verified as true and valid, its confidence or weight is increased accordingly; if a relationship is repeatedly disproven, its weight is decreased or it is even deleted. When new fault modes or maintenance experiences are discovered, the system adds them to the knowledge graph through knowledge extraction and fusion processes, ensuring the freshness and comprehensiveness of the knowledge.

[0048] The inference model optimization unit monitors and iteratively optimizes the algorithm models in the fault tracing and inference module. By comparing the root causes recommended by the system with those verified through actual maintenance, the system calculates metrics such as accuracy, recall, and F1 score. Supervised learning or reinforcement learning techniques (by rewarding successful repair cases to optimize the inference path selection strategy, or by penalizing erroneous inferences to adjust model parameters) are used to dynamically adjust inference rules and weight parameters to continuously improve the accuracy of future tracing. The fault detection model optimization unit periodically retrains and fine-tunes the machine learning model in the fault event detection module based on new operational data and fault labeling data to adapt to feature drift caused by equipment aging, process adjustments, and changes in raw materials, ensuring the robustness of the detection model. The performance monitoring and reporting unit continuously monitors the overall system's operational status, data processing volume, response time, inference success rate, and other key performance indicators, generating periodic reports to provide data-driven support for system maintenance and upgrades.

[0049] The present invention proposes a method for tracing the faults of papermaking equipment based on process knowledge graphs. The overall process is as follows: Figure 2 As shown, this method includes data acquisition and preprocessing, construction of a papermaking process knowledge graph, fault event detection and matching, fault tracing and propagation path reasoning, and maintenance solution recommendation and optimization. The method includes the following steps: S1, Data Acquisition and Preprocessing: Construct a multi-source data acquisition network to collect in real time the operating status data, process parameter data, production quality data, environmental parameter data, and historical fault and maintenance data of each core and auxiliary equipment during the papermaking process; The collected raw data includes, but is not limited to, operating status data such as refiner vibration signals, press temperature, dryer pressure, pulp flow rate, motor current, paper machine speed, and plant noise intensity; process parameter data such as pulp concentration, pH value, sizing degree, fiber length, filler content, white water turbidity, wire section vacuum degree, press line pressure, dryer cylinder surface temperature, dryer steam pressure, and paper web tension; and production quality data such as paper basis weight, thickness, moisture content, brightness, tensile strength, tear strength, and uniformity. Historical fault and maintenance data includes fault codes, fault description text, occurrence time, duration, maintenance operation records, replaced parts information, and maintenance personnel feedback. The collected raw data undergoes preprocessing operations such as cleaning, standardization, missing value imputation, outlier handling, and time series alignment, and is then uniformly stored in an integrated data warehouse. Data cleaning employs median filtering and wavelet denoising techniques based on sliding windows, while clustering-based outlier detection algorithms identify and correct outliers. Missing value imputation uses interpolation or regression prediction methods depending on the data type and missing value pattern. Time series alignment achieves synchronization of data at different frequencies through precise timestamp matching and resampling techniques. The integrated data warehouse is built on a distributed storage architecture, employing columnar storage database technology for efficient reading and writing of time series data, while using a relational database to store structured metadata and integrating a document-oriented database to store unstructured logs. S2, Papermaking Process Knowledge Graph Construction: Entities and relationships are extracted from structured and unstructured data sources, and an ontology of the papermaking process domain is constructed, thereby building a papermaking process knowledge graph. The structured data sources include equipment parameter manuals, process flow diagrams, SCADA system configuration data, historical fault code tables, and fault log databases, etc., and knowledge extraction is performed through predefined parsers and mapping rules. The unstructured data sources include maintenance reports, operation manuals, transcripts of expert experience interviews, technical specification documents, and industry standard texts, etc., and knowledge extraction is achieved through natural language processing technology and deep learning models. The named entity recognition module is based on a pre-trained language model with a Transformer architecture, and is fine-tuned by combining conditional random field layers or pointer networks. It can accurately identify entity types in the text, such as equipment parts, fault types, fault phenomena, fault causes, process parameters, maintenance measures, and required tools and consumables. The relation extraction module employs graph neural networks or deep learning models based on attention mechanisms to learn the association patterns between entities. The event extraction module identifies key elements in the occurrence, development, and handling of fault events, structuring them into event entities and event attributes. For example, it identifies the main equipment involved in a paper web breakage event, the time of occurrence, related phenomena, possible causes, and the handler. The papermaking process ontology defines all core concepts, attributes, and their interrelationships in the papermaking production process, providing a unified semantic framework to guide the construction and integration of the knowledge graph, ensuring the standardization and consistency of knowledge. The papermaking process knowledge graph is stored in a graph database and adopts an attribute graph model. Each node has a unique identifier and a set of attributes, and each edge represents a specific relationship type and can carry attribute descriptions of the relationship features. Figure 3 This invention presents a schematic diagram illustrating the structure of the papermaking process knowledge graph. S3, Fault Event Detection and Matching: Continuously monitors real-time operational data in the integrated data warehouse and detects equipment anomalies and fault events through a pre-set fault diagnosis model. When a fault event is detected, the system automatically extracts its key features and matches them with known fault patterns in the knowledge graph; The fault diagnosis model includes: a control chart method based on statistical process control, used to monitor whether the mean and fluctuation of key process parameters exceed control limits; an anomaly detection algorithm based on machine learning, used to identify nonlinear anomaly patterns in multivariate time-series data; and a rule-based expert system, used to identify composite anomalies that meet specific logical conditions. Once a fault event is detected, the system automatically extracts information such as the event's occurrence time, involved equipment, degree of abnormal deviation of associated parameters, duration of abnormality, and preliminary symptom description, constructing a structured fault event feature vector. The matching process is implemented by calculating the similarity between the fault event feature vector and standardized fault pattern entities stored in the knowledge graph. A vector similarity matching method based on semantic embedding is used, which maps entities and relationships to a low-dimensional vector space and measures the matching degree by calculating cosine similarity. Alternatively, graph query algorithms based on attribute and relation pattern matching, such as performing Cypher or Gremlin queries in a graph database, can be used to find the fault pattern subgraph that is most similar to the attributes and relational structure described in the fault event report. The matching results are a set of fault type nodes, fault phenomenon nodes, and preliminary cause nodes that are most relevant to the current fault event, along with a similarity score. S4, Fault Origin and Propagation Path Reasoning: Starting from the matched fault type or fault phenomenon node, perform association queries and logical reasoning in the papermaking process knowledge graph, combine real-time data and historical case data, traverse the graph along the semantic relationship edges, gradually trace to the potential root cause of the fault, and construct the fault propagation path. Figure 4 A schematic diagram illustrating the fault tracing and propagation path reasoning of this invention is shown. The associated query and logical reasoning process includes: Initial root cause exploration: Starting from the initial fault phenomenon node, a graph traversal algorithm is used to explore upstream in the knowledge graph along causal relationship edges such as cause, origin, trigger, and impact to find the direct cause node. During the exploration process, the exploration path is pruned or weighted based on the abnormal correlation of parameters in real-time data. If real-time data shows that the reading of an upstream steam pressure sensor is consistently abnormally low, the potential cause path related to that steam pressure will be given higher priority. Similarly, if the reading of the condensate return flow sensor is also abnormal, the path weight for poor condensate return will also increase. This real-time data-driven path evaluation mechanism significantly improves the accuracy of source tracing. Multi-level causal chain construction: For the identified direct cause nodes, recursive tracing continues, following the relationships such as "caused by," "is the cause of," "affects," "inadequate maintenance," and "operational errors," to delve deeper into the root causes. During this process, the system comprehensively considers the weight and priority of different types of relationship edges. Equipment component failures are usually direct causes, while process parameter deviations or raw material problems may be deeper root causes, and management or maintenance issues may be the deepest root causes. The system utilizes graph pattern matching technology to identify common fault propagation patterns and uses them as predefined inference rules to accelerate the tracing process. Historical Case Validation and Supplementation: Potential fault paths and root causes formed during the reasoning process are compared and validated with historical fault cases in the integrated data warehouse. If a historical case highly similar to the current reasoning result exists, the validated fault propagation path and maintenance experience from that case are used to strengthen or correct the current reasoning result. Simultaneously, the system uses data mining and machine learning techniques to extract new, unmodeled causal relationships or association patterns from historical cases and incrementally updates them to the knowledge graph to continuously improve its completeness and accuracy. Multi-source information fusion reasoning: During the reasoning process, the system dynamically integrates real-time sensor data with structured knowledge from the knowledge graph. When the knowledge graph infers that poor bearing lubrication is a potential cause, the system will further query real-time temperature and vibration sensor data. If the bearing temperature continues to rise and vibration spectrum analysis shows an increase in energy at the bearing fault characteristic frequency, the reliability of the reasoning is further confirmed. Conversely, if the real-time data is inconsistent, the reasoning direction is adjusted or the priority of that path is reduced. This real-time feedback mechanism ensures the accuracy and dynamic adaptability of the reasoning results. Root Cause Prioritization: For all identified potential root causes, a comprehensive evaluation is conducted based on their centrality in the knowledge graph, their matching degree with historical failure cases, the confidence level of real-time data validation, and the severity of their impact on production. A priority is assigned to determine the most likely set of root causes. The evaluation employs a multi-criteria decision analysis method, providing a quantified priority score for each root cause. Fault Propagation Path Construction and Visualization: Based on the reasoning results, a complete fault propagation path from the fault phenomenon to the ultimate root cause is constructed. This path consists of a series of nodes connected by semantic relationships. The propagation path is presented graphically on the user interface, including key nodes, and different types of relationships or levels of importance can be distinguished by different colors or thicknesses. The user interface supports zooming in and out of the path, querying node information, and path animation demonstrations, so that maintenance personnel can intuitively understand the occurrence and development process of the fault and assist in decision-making. S5, Maintenance Solution Recommendation and Optimization: Based on the identified root causes and propagation paths of the fault, the system performs a correlation query in the papermaking process knowledge graph to extract maintenance experience, maintenance operating procedures, required tools and spare parts information related to the fault type, equipment model, root cause, and historical maintenance records. It then automatically generates a structured maintenance solution recommendation. This recommendation includes, but is not limited to: detailed fault handling steps, necessary safety precautions, a recommended list of maintenance tools, required spare parts models and quantities, estimated maintenance time, and relevant expert advice. The system further provides a user feedback interface, allowing maintenance personnel to input the actual maintenance results into the system after completing the maintenance. The system continuously optimizes and iterates the relation weights, inference rules, and maintenance plan recommendation models in the knowledge graph through reinforcement learning or supervised learning algorithms, combined with new feedback data, thereby improving the accuracy and effectiveness of future fault tracing and maintenance plan recommendations.

[0050] The optimization mechanism includes: strengthening the confidence of corresponding causal relationships in the knowledge graph based on successful repair cases; analyzing the causes of failure and correcting relevant knowledge or reasoning rules based on failed repair cases; and adjusting the priority strategy of the repair plan recommendation algorithm by comparing the effects of recommended solutions with actual implementation solutions. If the system recommends solution A, although technically feasible, but the repair time is too long, while solution B used in actual repair is more efficient, the system will adjust the recommendation priority of solution A or learn the advantages of solution B.

[0051] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for tracing the source of faults in papermaking equipment based on process knowledge graphs, characterized in that, Includes the following steps: S1, Data Acquisition and Preprocessing: Construct a multi-source data acquisition network to collect real-time operating status data, process parameter data, production quality data, environmental parameter data, and historical fault and maintenance data of each core and auxiliary equipment during the papermaking process; perform cleaning, standardization, missing value filling, outlier handling, and time series alignment preprocessing operations on the collected raw data, and store the processed data in an integrated data warehouse. S2, Construction of Papermaking Process Knowledge Graph: Extract entities and relationships from structured and unstructured data sources, construct an ontology of the papermaking process domain, and then construct a papermaking process knowledge graph; the papermaking process knowledge graph uses nodes to represent entities in the papermaking production process and edges to represent semantic relationships between entities; S3, Fault Event Detection and Matching: Continuously monitor the real-time operating data in the integrated data warehouse, and detect equipment anomalies and fault events through a preset fault diagnosis model; when a fault event is detected, extract its key features and match them with known fault modes in the papermaking process knowledge graph; S4, Fault Origin Tracing and Propagation Path Reasoning: Starting from the matched fault type or fault phenomenon node, perform association queries and logical reasoning in the papermaking process knowledge graph, including: Starting from the initial fault phenomenon node, the knowledge graph is explored upstream using a graph traversal algorithm to find the direct cause node; at the same time, the exploration path is pruned or weighted based on the abnormal correlation of parameters in real-time data. For the identified direct cause nodes, continue to perform recursive tracing; The potential failure paths and root causes formed during the reasoning process are compared and verified with historical failure cases in the integrated data warehouse; During the reasoning process, the real-time data and the structured knowledge in the knowledge graph are dynamically integrated, and the real-time data is used as a constraint or verification evidence for the reasoning. For all identified potential root causes, a comprehensive evaluation is conducted and priorities are assigned based on their centrality in the knowledge graph, their matching degree with the historical failure cases, the confidence level of the real-time data verification, and the severity of their impact on production, to determine the root cause set. Based on the reasoning results, a complete fault propagation path from the fault phenomenon to the final root cause is constructed, and the propagation path is presented in a graphical manner on the user interface; S5, Maintenance Solution Recommendation and Optimization: Based on the identified root causes and propagation paths of the faults, a correlation query is performed in the papermaking process knowledge graph to extract maintenance experience, maintenance operating procedures, required tools and spare parts information related to the fault type, equipment model, root cause, and historical maintenance records, and to automatically generate structured maintenance solution recommendations; the method also includes: further providing a user feedback interface, and continuously optimizing and iterating the relation weights, inference rules, and maintenance solution recommendation model in the papermaking process knowledge graph based on the actual maintenance results of the maintenance personnel.

2. The method for tracing the source of papermaking equipment faults based on process knowledge graphs according to claim 1, characterized in that, The multi-source data acquisition network is achieved by deploying various industrial sensors and data acquisition terminals. The sensors are interconnected with the edge computing gateway and host computer system via industrial Ethernet, wireless sensor network or fieldbus to achieve high-frequency, low-latency data transmission; The collected operational data includes vibration signals, temperature, pressure, flow rate, current, voltage, rotational speed, and noise intensity. The process parameter data includes pulp concentration, pH value, sizing degree, fiber length, filler content, white water turbidity, wire section vacuum degree, press line pressure, drying cylinder surface temperature, drying section steam pressure, and paper web tension. The production quality data covers paper basis weight, thickness, moisture content, whiteness, tensile strength, tear resistance, and uniformity. The historical fault and maintenance data includes fault codes, fault description text, occurrence time, duration, maintenance operation records, information on replaced parts, and feedback from maintenance personnel.

3. The method for tracing the source of papermaking equipment faults based on process knowledge graphs according to claim 2, characterized in that, The preprocessing operations for cleaning, standardizing, filling missing values, handling outliers, and aligning time series data include, but are not limited to: the data cleaning uses median filtering based on sliding window, wavelet denoising technology, and clustering-based outlier detection algorithm to identify and correct outliers. The missing value imputation uses interpolation or regression prediction methods based on the data type and missing value pattern. The time series alignment achieves synchronization of data at different frequencies through precise timestamp matching and resampling techniques. The integrated data warehouse is built on a distributed storage architecture, uses columnar storage database technology for efficient reading and writing of time-series data, uses a relational database to store structured metadata, and integrates a document database to store unstructured logs.

4. The method for tracing the source of papermaking equipment faults based on process knowledge graphs according to claim 1, characterized in that, The knowledge extraction process in step S2 is implemented using natural language processing technology and deep learning models, including named entity recognition, relation extraction and event extraction. The named entity recognition module is based on a pre-trained language model of the Transformer architecture, and is fine-tuned by combining a conditional random field layer or a pointer network to identify equipment components, fault types, fault phenomena, fault causes, process parameters, maintenance measures, tools and consumables entities in the text. The relationship extraction module uses graph neural networks or deep learning models based on attention mechanisms to learn the association patterns between entities. The papermaking process ontology defines all core concepts, attributes, and their interrelationships in the papermaking production process, providing a unified semantic framework; the papermaking process knowledge graph is stored in a graph database and uses an attribute graph model.

5. The method for tracing the source of papermaking equipment faults based on process knowledge graphs according to claim 1, characterized in that, The fault diagnosis model includes, but is not limited to: a control chart method based on statistical process control, used to monitor the mean and volatility of key process parameters; Machine learning-based anomaly detection algorithms are used to identify nonlinear anomaly patterns in multivariate time series data; rule-based expert systems are used to identify complex anomalies. The matching process is achieved by calculating the similarity between the fault event feature vector and the standardized fault pattern entities stored in the knowledge graph, using a vector similarity matching method based on semantic embedding or a graph query algorithm based on attribute and relation pattern matching.

6. A fault tracing system for papermaking equipment based on process knowledge graphs, characterized in that, The method for tracing the source of papermaking equipment faults based on process knowledge graphs, as described in any one of claims 1-5, includes: The data acquisition module is used to configure multi-channel data acquisition units and various industrial sensor arrays to collect various operating data on the paper production line in real time. The data preprocessing and storage module is used to receive the raw data transmitted by the data acquisition module, and to perform data cleaning, standardization, data fusion and storage on it; The knowledge graph construction and management module is used for the creation, updating, and maintenance of the papermaking process knowledge graph; The fault event detection module is used to analyze the data in the data preprocessing and storage module in real time to identify abnormal equipment operation or fault events. The fault tracing and reasoning module is used to perform in-depth fault causal reasoning using the papermaking process knowledge graph. The maintenance decision support and visualization module is used to present the fault tracing results to the user in an intuitive way and provide intelligent maintenance solution recommendations. The system feedback and optimization module is used for continuous learning and optimization of the entire system.

7. The system according to claim 6, characterized in that, The data acquisition module includes: the industrial sensor array including but not limited to piezoelectric accelerometers, resistance temperature sensors, capacitive pressure sensors, electromagnetic flowmeters, clamp-on current transformers, photoelectric speed sensors, MEMS microphones and industrial vision cameras installed in key parts of various papermaking equipment. The data acquisition unit integrates an analog-to-digital converter, a signal conditioning circuit, and a data buffer, and transmits data with the data preprocessing and storage module through an industrial Ethernet interface or a wireless communication module. The data acquisition module supports multiple communication protocols. The data preprocessing and storage module includes: the data cleaning function includes median filtering based on sliding window, wavelet denoising, and outlier detection algorithm based on machine learning; the data standardization function includes Min-Max standardization or Z-score standardization. The data fusion function integrates data from different data sources through timestamp alignment algorithms and semantic association algorithms based on entity IDs; the storage unit includes high-performance time-series databases, relational databases, and unstructured data storage.

8. The system according to claim 7, characterized in that, The knowledge graph construction and management module includes: a knowledge extraction unit, used to perform named entity recognition using a pre-trained deep learning model to identify entity types in the papermaking field, and to extract relationships using a model based on graph neural networks or multi-head attention mechanisms to identify semantic relationships between entities; the structured data parser maps structured information into nodes and edges of the knowledge graph; The ontology construction unit is used to provide domain ontology modeling tools, define the core concepts, attributes and relational patterns of the papermaking production domain, and ensure the semantic consistency and scalability of the knowledge graph. The knowledge fusion unit is used to integrate knowledge from different sources and formats to resolve entity disambiguation and relationship conflict issues. The knowledge storage and query unit is used to store the constructed knowledge graph using a high-performance graph database and to provide an efficient graph query language interface. The knowledge update and maintenance unit provides a user interface for manual editing and proofreading of knowledge, and has an automatic or semi-automatic knowledge incremental update mechanism.

9. The system according to claim 8, characterized in that, The fault tracing and reasoning module includes: The starting node matching unit is used to receive the output of the fault event detection module and, based on the key information in the fault event report, match the most relevant fault phenomenon node or fault type node in the knowledge graph as the starting point for tracing the source. The graph traversal and path discovery unit is used to execute a customized graph traversal algorithm in the knowledge graph, starting from the matched starting node. The algorithm combines the characteristics of breadth-first search and depth-first search, prioritizing the exploration along edges with strong causal relationships that cause, trigger, or influence. It also considers path length, node importance, and edge weight to optimize the search direction and dynamically prunes paths that do not conform to real-time data logic or are unreasonable. A logic reasoning engine is used to embed a rule-based inference engine and a graph-based reasoning algorithm. The rule-based inference engine performs deductive reasoning based on predefined Datalog or Prolog rules, and the graph-based reasoning algorithm uses node and edge representations learned by graph neural networks or knowledge graph embedding models to perform link prediction or node classification. The multi-source information fusion reasoning unit is used to dynamically acquire real-time running data and historical case data related to the current reasoning path during the reasoning process, and use the real-time data as constraints or verification evidence for reasoning. The root cause identification and priority assessment unit is used to comprehensively evaluate all potential root causes obtained through reasoning, and calculate their confidence or priority. The assessment indicators include the reliability of the tracing path, the degree of consistency with real-time data, the corroboration strength of historical cases, the severity of the impact on production, and the maintenance cost. The fault propagation path construction unit is used to construct a complete directed graph structure from the fault phenomenon to the most likely root cause based on the results of the root cause identification and priority evaluation.

Images