An Industrial Fault Diagnosis Method and System Based on Natural Language Fault Features and Large Model Knowledge-Enhanced Reasoning

Abstract

This invention relates to the technical field of industrial equipment condition monitoring and intelligent fault diagnosis, and discloses an industrial fault diagnosis method and system based on natural language fault features and large-scale model knowledge-enhanced reasoning. The method includes: during equipment operation, acquiring multi-source monitoring and condition analysis results from the equipment under test; converting the features into structured natural language fault feature text; vectorizing the fault feature text and performing similarity retrieval in a pre-built industrial fault knowledge base to recall knowledge fragments; constructing reasoning prompts based on the fault feature text and recalled knowledge fragments, inputting them into a large language model for knowledge-enhanced reasoning, and generating diagnostic results. This invention effectively solves the problems of existing intelligent diagnostic technologies based on rule bases or knowledge graphs, such as difficulty in unifying numerical evidence and semantic knowledge, high costs of knowledge coverage and updates, and insufficient consistency of conclusions across operating conditions.

Description

Technical Field

[0001] This invention relates to the technical field of industrial equipment condition monitoring and intelligent fault diagnosis, and in particular to an industrial fault diagnosis method and system based on natural language fault features and large model knowledge-enhanced reasoning. Background Technology

[0002] Equipment health management (PHM) in industries such as manufacturing, energy, and transportation has gradually evolved from diagnosis based on expert experience / rule bases to an integrated framework of monitoring, analysis, and decision-making driven by multi-source signals. Traditional methods focus on single modes (such as vibration) or threshold judgments of a small number of statistics, making it difficult to achieve stable migration in complex operating conditions and across different equipment models. Although data-driven methods can extract features such as spectral peaks / octaves, shaft center trajectory shapes, and trend slopes from signals such as vibration, temperature, and current, they are still insufficient in terms of knowledge reuse, interpretable reasoning, and traceable evidence.

[0003] In recent years, the capabilities of Large Language Models (LLM) in natural language understanding, knowledge-based question answering, and causal reasoning have provided a new path for closing the "data-knowledge-reasoning" loop in industry. However, LLM is natively text-based and has limited ability to analyze numerical / shape-based evidence such as spectrum, trajectory, and trend. Furthermore, it lacks a systematic knowledge modeling and enhancement mechanism for the industrial field, resulting in shortcomings in evidence constraint, cross-condition generalization, and conclusion consistency when directly used for high-reliability diagnosis. Therefore, there is an urgent need for a technical route that can connect multi-source numerical evidence → standardized semantic representation → domain knowledge enhancement → interpretable reasoning diagnosis to achieve robust diagnosis and traceable decision-making for rotating equipment such as pumps, compressors, fans, and motors under real-world operating conditions.

[0004] CN118312896A discloses a fault diagnosis expert system based on a large language model. It uses a rule base expressed in natural language as its core, extracts rule features by LLM, and then uses sensors to calculate corresponding values ​​and make judgments according to the rules. This scheme is beneficial to alleviate the hard coding of rules, but its diagnostic link is still essentially rule → value verification. It does not establish a unified transcription mechanism from numerical or shape evidence such as spectrum / axis trajectory / trend to standardized natural language fault features, nor does it disclose the Top-K knowledge retrieval and evidence splicing prompt word process based on vector similarity for equipment type-component-operating condition context. It is difficult to maintain robustness and traceability under cross-equipment and complex operating conditions.

[0005] CN117271779A discloses a fault analysis method combining a large model and a knowledge graph (KG). The method uses a large model + knowledge graph (KG) paradigm: keywords are extracted from the user's natural language input, and the fault type is determined by KG adjacency retrieval, and a solution is given. This method is friendly to text scenarios, but it lacks support for structured access and semantic alignment of multi-source numerical signals. The construction and maintenance of KG is costly and the granularity is limited. It also lacks vectorized cross-document recall and structured output constraints / consistency checks and rollback mechanisms for unstructured materials such as manuals / cases / experiences.

[0006] In summary, existing intelligent diagnostic technologies based on rule bases or knowledge graphs generally suffer from problems such as difficulty in unifying numerical evidence and semantic knowledge, high costs of knowledge coverage and updating, and insufficient consistency of conclusions across operating conditions. This invention proposes a systematic solution to the key problem of semanticizing multi-source numerical / shape fault features and integrating them with a vectorized knowledge base to drive knowledge-enhanced reasoning in large models. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: during the equipment operation phase, multi-source monitoring and status analysis results from the tested equipment are obtained, and the analysis results include at least vibration spectrum characteristics, shaft center trajectory characteristics and operating parameter trend characteristics;

[0010] The features are then converted into fault feature text in structured natural language;

[0011] The fault feature text is vectorized and encoded, and a similarity retrieval is performed in a pre-built industrial fault knowledge base to recall knowledge fragments related to equipment type, components and operating conditions.

[0012] Based on the fault feature text and recalled knowledge fragments, reasoning prompts are constructed and input into a large language model for knowledge-enhanced reasoning to generate diagnostic results that include fault type, affected components, diagnostic basis, and handling suggestions.

[0013] As a preferred embodiment of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning described in this invention, the step of acquiring multi-source monitoring and condition analysis results from the equipment under test includes:

[0014] A fast Fourier transform is performed on the original vibration signal, representative spectral peaks are extracted based on a preset peak prominence threshold, and similar peaks are merged according to frequency resolution to determine the fundamental frequency and its harmonics.

[0015] The X and Y direction displacements are collected to form a set of axis trajectory points. The centroid is translated and aligned with the main direction. The curvature perturbation and distance perturbation are calculated and the number of trajectory intersections is detected. Based on this, the trajectory is determined to be a figure-eight shape, a flat ellipse, a banana shape, or a normal ellipse.

[0016] Within a sliding time window, the mean, standard deviation, and trend slope of slowly varying parameters such as temperature, current, and vibration amplitude are calculated. Combined with dynamic threshold intervals, out-of-limit, fluctuation anomaly, and trend anomaly are determined to form structured state analysis results.

[0017] As a preferred embodiment of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning described in this invention, the generation of the fault feature text includes:

[0018] Extract the main peak frequency, octave / sideband interval, and amplitude ratio from the spectral features, and generate sentences according to preset word patterns;

[0019] The trajectory shape is obtained from the axis trajectory features, and sentences are generated according to preset word patterns;

[0020] The mean, volatility, and trend slope of the parameters are calculated, and the quantification results are mapped to level labels and synthesized into a multi-sentence structured text description.

[0021] As a preferred embodiment of the industrial fault diagnosis method based on natural language fault features and large-model knowledge-enhanced reasoning described in this invention, the step of performing similarity retrieval in a pre-built industrial fault knowledge base to recall knowledge fragments related to equipment type, components, and operating conditions includes:

[0022] The fault feature text is encoded into a query vector using an embedding model;

[0023] Unstructured knowledge from maintenance manuals, historical cases, and expert experience is segmented and encoded according to equipment type, faulty component, and fault characteristics to create a vector index;

[0024] Top-K search is performed based on cosine similarity, and filtered by device model, operating condition label and timestamp to return the knowledge fragments and their source identifiers that are most relevant to the current scenario.

[0025] As a preferred embodiment of the industrial fault diagnosis method based on natural language fault features and large-scale model knowledge-enhanced reasoning described in this invention, the industrial fault knowledge base includes structured data and unstructured documents, wherein:

[0026] Structured data includes parameter threshold tables, component hierarchies, and common fault cause tables;

[0027] Unstructured documents include maintenance manuals, historical failure cases, and expert experience entries;

[0028] Unstructured documents are segmented by paragraphs with identifiers and encoded into vectors by an embedding model. Vector indexes are then built to support Top-K retrieval and filtering based on device model, operating condition tags, and timestamps.

[0029] As a preferred embodiment of the industrial fault diagnosis method based on natural language fault features and large-scale model knowledge-enhanced reasoning described in this invention, reasoning prompts are constructed based on the fault feature text and recalled knowledge fragments, and input into a large language model for knowledge-enhanced reasoning, including:

[0030] Construct reasoning prompts based on prompt templates for roles, tasks, constraints, and formats;

[0031] The fault feature text is concatenated with Top-K knowledge fragments and input into a large language model. The fault type, fault manifestation, fault analysis and reasoning process, diagnostic result confidence and maintenance suggestions are output according to structured fields.

[0032] When the matching degree between the fault feature text and the Top-K knowledge fragment is lower than the threshold, the fault type output is no fault, the diagnostic result confidence output is low, and the remaining fields are left blank.

[0033] As a preferred embodiment of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning described in this invention, the construction of the reasoning prompt words adopts a four-element template of role, task, constraint, and format, wherein the four-element template includes at least:

[0034] The role is defined as an industrial fault diagnosis expert;

[0035] The task is to perform causal analysis by combining fault feature text with knowledge fragments and to provide actionable handling suggestions.

[0036] The constraints include uniformity of units, standardization of component names, and traceability of conclusions to specific evidence items.

[0037] The format requires output in structured fields.

[0038] As a preferred embodiment of the industrial fault diagnosis system based on natural language fault features and large model knowledge-enhanced reasoning described in this invention, it includes:

[0039] The multi-source sensor module is used to collect signals such as vibration, temperature, sound and electrical quantities of the device under test, and output them through a standard industrial interface;

[0040] The data acquisition and analysis unit is electrically connected to the multi-source sensor module and is used to perform signal conditioning, analog-to-digital conversion, filtering and standardization on the acquired signals and generate state analysis results and fault characteristic data.

[0041] The fault diagnosis reasoning unit is communicatively connected to the data acquisition and analysis unit, and is used to transcribe the fault feature data into fault feature text, perform vector retrieval on the fault knowledge base and perform large model reasoning diagnosis, and output structured diagnosis results.

[0042] The results display service module is communicatively connected to the fault diagnosis reasoning unit and is used to receive and provide diagnostic results, evidence fragments, and original / derived data in a service manner.

[0043] The display unit is used to display the spectrum, axis trajectory, parameter trends, alarm information and the structured diagnostic results in a graphical interface, and supports interactive querying.

[0044] The external communication and data interface module is used to realize data interaction and remote synchronization with the host computer or PLC system;

[0045] The power management module is used to supply power to the above-mentioned units and modules and provide overvoltage, undervoltage and overcurrent protection.

[0046] As a preferred embodiment of the industrial fault diagnosis system based on natural language fault features and large model knowledge-enhanced reasoning according to the present invention, the system further includes one or more processors;

[0047] The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning as described above.

[0048] As a preferred embodiment of the computer-readable medium for the storage software described in this invention, the software includes instructions executable by one or more computers, the instructions causing the one or more computers to perform operations, the operations including the flow of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning as described above.

[0049] The beneficial effects of this invention are as follows: By using structured acquisition and textual expression of multi-source data, this invention improves the transferability and traceability of evidence across operating conditions and machine models; by using vector retrieval and tag filtering, it effectively incorporates unstructured manuals / cases into the reasoning context, alleviating the problems of high knowledge coverage and update costs; by using controlled prompts and structured output, it strengthens the causal chain and evidence citation, reduces risks, and facilitates on-site execution; at the same time, it achieves full-link interface standardization, which facilitates edge deployment and system integration, has good maintenance convenience, and significantly improves the reliability of diagnostic decisions. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0051] Figure 1 This is a flowchart illustrating the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning as shown in this invention.

[0052] Figure 2 This is an example diagram of spectrum data shown in the present invention;

[0053] Figure 3 This is a diagram of the axis trajectory shown in this invention;

[0054] Figure 4 This is a schematic diagram of the module structure distribution of the industrial fault diagnosis system based on natural language fault features and large model knowledge-enhanced reasoning as shown in this invention. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0056] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0057] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0058] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates an industrial fault diagnosis method based on natural language fault features and large-scale model knowledge-enhanced reasoning, which specifically includes the following steps:

[0059] S1. During the equipment operation phase, acquire multi-source monitoring and status analysis results from the equipment under test. The analysis results shall include at least vibration spectrum characteristics, shaft trajectory characteristics, and operating parameter trend characteristics.

[0060] S2. Convert the features into structured natural language fault feature text;

[0061] S3. Vectorize the fault feature text and perform similarity retrieval in the pre-built industrial fault knowledge base to recall knowledge fragments related to equipment type, component and operating condition.

[0062] S4. Construct reasoning prompts based on fault feature text and recalled knowledge fragments, input them into a large language model for knowledge-enhanced reasoning, and generate diagnostic results that include fault type, affected components, diagnostic basis, and handling suggestions.

[0063] It should be noted that, compared with the prior art, the diagnostic method provided in this embodiment of the invention structures the results of multi-source monitoring and state analysis (spectral features, shaft trajectory features, parameter trend features) into natural language fault feature text, then performs vectorization encoding and performs similarity retrieval in a pre-built industrial fault knowledge base to recall knowledge fragments related to equipment type, components, and operating conditions. Subsequently, it drives LLM to complete knowledge-enhanced reasoning with role-task-constraint-format prompt word templates, and outputs structured conclusions containing fault type, affected components, diagnostic basis, and handling suggestions, thereby establishing a transferable, verifiable, and traceable closed loop between numerical evidence, semantic knowledge, and model reasoning.

[0064] Preferably, step S1 acquires and analyzes multiple types of signals simultaneously during the equipment operation phase to form structured features covering the frequency domain, trajectory morphology, and slowly varying parameters, thereby achieving a comprehensive characterization and evidence retention of fault symptoms; thus reducing the risk of missing a single signal, improving the sensitivity and stability of early anomaly identification, and providing clear and traceable basic data for subsequent semantic processing and reasoning.

[0065] Preferably, step S2 transcribes the quantification results such as spectral peak, overtone / sideband, trajectory shape, and parameter trends into a controlled vocabulary and a unified sentence structure for fault feature text, thereby achieving consistency and comparability of feature expressions under different devices and operating conditions. This facilitates subsequent retrieval and model consumption, reduces ambiguity caused by differences in manual descriptions, and retains the correspondence with the original data, making it easier to review and track.

[0066] Preferably, step S3 involves vectorizing the fault feature text and performing a similarity search in an industrial fault knowledge base organized by equipment type, component, and operating condition to achieve rapid association with historical cases, maintenance manuals, and expert experience. This improves the relevance and relevance of the search results to the current scenario, reduces interference from irrelevant information, shortens the time for locating the cause and finding handling suggestions, and provides citationable evidence fragments for subsequent reasoning.

[0067] Preferably, step S4 involves inputting the fault feature text and the recalled knowledge fragment into the large model according to the preset prompt word template, and outputting a structured result containing the fault type, affected components, diagnostic basis and treatment suggestions, thereby realizing causal chain reasoning and interpretable conclusion generation with the support of external knowledge, thereby improving the accuracy and consistency of diagnosis.

[0068] The following is combined with Figure 2 , Figure 3 The schematic diagrams shown, along with some preferred or optional examples of the present invention, more specifically describe the implementation process and / or effects of certain embodiments of the present invention.

[0069] It should be noted that, in order to convert structured fault features into natural language descriptions, three types of key features from the state analysis module are used as input data: vibration spectrum features, shaft center trajectory features, and parameter trend features. Their data formats are as follows:

[0070] Vibration spectrum:

[0071] By analyzing the raw signals of vibration velocity or acceleration The frequency domain amplitude spectrum is obtained by performing a Fast Fourier Transform (FFT). Its structure is a set of key-value pairs of frequencies and corresponding amplitudes, that is:

[0072]

[0073] in, Indicates the frequency axis. This is the corresponding amplitude sequence.

[0074] Axis trajectory:

[0075] The two-dimensional motion trajectory within the rotor's cross-sectional plane is obtained using a dual-channel axial displacement sensor, and the sampled trajectory point sequence is represented as follows:

[0076]

[0077] in, For the first The horizontal displacement of each sampling point. For the first The vertical displacement of each sampling point forms a continuous trajectory graphic. After normalization, it becomes a standard coordinate point array input, which is used by the shape recognition model to determine the trajectory category (such as ellipse, figure-eight, eccentricity, etc.). The input form is a set of two-dimensional vectors, and no additional label information is required.

[0078] Parameter characteristics:

[0079] It should include at least slowly changing monitoring signals such as temperature, current, and vibration amplitude, and record their changes in the form of a time series. The following key statistics were extracted as input:

[0080] average value:

[0081] Volatility (standard deviation):

[0082] Trend slope (least squares method):

[0083] in, To be at the sampling time The monitored parameter values ​​(such as temperature, current, or vibration amplitude). For the first The time point of the next sampling This represents the total number of sampling points.

[0084] Furthermore, a fault feature text description that can be understood by a large language model is generated through a feature generation model, serving as input for subsequent diagnostic reasoning. This includes the generation of vibration spectrum fault feature descriptions, shaft center trajectory fault feature descriptions, and parameter anomaly descriptions.

[0085] (1) Generation of vibration fault feature description

[0086] Input data graph example Figure 2 As shown, vibration velocity or acceleration signals are collected under operating conditions, and spectral data are obtained through Fast Fourier Transform (FFT), which is represented as a mapping relationship between a set of amplitude sequences and the corresponding frequency axes:

[0087]

[0088] in, Indicates the frequency axis. For the corresponding amplitude sequence;

[0089] To extract key spectral features, the salience threshold for extracting significant peaks is first calculated based on the dynamic range of the spectral amplitude. :

[0090]

[0091] in, This is an adjustable coefficient (default value is 0.2) used to control peak sensitivity;

[0092] Based on this, a spectrum peak search is performed to identify all peaks that meet the salience requirement greater than [value missing]. The peak point is denoted as:

[0093]

[0094] in, For the first The frequencies corresponding to each spectral peak For the set of all peak points;

[0095] To reduce noise impact and merge similar frequency components, a similar peak merging rule is introduced, with the frequency spacing filter setting set as follows:

[0096]

[0097] in, Frequency resolution;

[0098] Frequency spacing less than Multiple peaks are combined to obtain their mean frequency, thus yielding the final representative peak set. ,in For the merged first A representative peak frequency, This represents the final number of representative peaks;

[0099] The first frequency peak is taken as the fundamental frequency by default. The remaining frequency peaks are classified according to their ratio to the fundamental frequency:

[0100] like ,in Then it is considered as Frequency multiplication;

[0101] The remaining frequencies are recorded as "other frequency peaks";

[0102] Generate text describing the spectral features of natural language structures, for example:

[0103] Output begins: "Fundamental frequency of a vibration measuring point: 100Hz, second harmonic: 200Hz, other frequency peaks: 1000Hz."

[0104] The vibration spectrum feature description is used as input to the inference model to assist in subsequent fault causal analysis and diagnostic conclusion generation.

[0105] (2) Generation of shaft center trajectory fault feature description

[0106] The input example is a graph.Figure 3 As shown, during equipment operation, axial displacement signals in the X and Y directions are collected to construct a two-dimensional trajectory point set:

[0107]

[0108] in, For the first The horizontal displacement of each sampling point. For the first The vertical displacement of each sampling point;

[0109] To enhance the generality and robustness of the analysis, the trajectory is first normalized, including centroid translation and principal direction alignment.

[0110] Calculate the coordinates of the trajectory center (centroid):

[0111]

[0112] The trajectory points after translation are:

[0113]

[0114] Principal component analysis (PCA) is performed on the trajectory points, and the variance ratio of the first principal component to the second principal component is extracted as a criterion for the flattening of the ellipse.

[0115]

[0116] in, The eigenvalues ​​corresponding to the main direction;

[0117] Calculate the curvature sequence of the translated trajectory:

[0118]

[0119] in, No. The curvature of each sampling point is used to measure the degree of curvature of the trajectory at that point. Let be the first derivative of the translated trajectory in the x and y directions, and represent the velocity components. Let x be the second derivative of the trajectory after translation in the x and y directions, and y be the acceleration components.

[0120] Extract the extreme values ​​and median of curvature, and define the curvature perturbation ratio:

[0121]

[0122] in, Curvature of all sampling points A set of sequences;

[0123] The Euclidean distance from each point to the centroid of the trajectory is:

[0124]

[0125] Calculate its mean and standard deviation to obtain the distance perturbation ratio:

[0126]

[0127] in, The Euclidean distance from all sampling points to the centroid of the trajectory A set of sequences;

[0128] For the trajectory at multiple rotation angles The process involves reprojecting the line segments to detect any intersections and counting the maximum number of intersections. The specific process is as follows:

[0129] For any rotation angle The trajectory points are rotated around the origin in two dimensions to obtain the rotated coordinates:

[0130]

[0131] Define the set of line segments formed by continuous trajectory points:

[0132]

[0133] For any two non-adjacent line segments (line segments) , Determine whether the line segments meet the intersection condition (using the cross product):

[0134] judge:

[0135] The total number of intersections at this angle is ;

[0136] From multiple angles (equally spaced) (Inner) Repeat the above rotation and intersection detection operations, and finally take the maximum number of intersection points as one of the shape features of the trajectory:

[0137]

[0138] Based on the above multiple shape feature indicators, the trajectories are classified into the following typical types, and natural language description results are generated:

[0139] like It is identified as a figure-eight shape.

[0140] like and and It is identified as: flat oval shape;

[0141] like and and It is identified as: banana-shaped;

[0142] Otherwise, it will be identified as a normal oval shape.

[0143] For example, the final axis trajectory shape description output is:

[0144] Output begins: "The axis trajectory is in a figure-eight shape";

[0145] Output begins: "The axis trajectory is a flattened ellipse";

[0146] Output begins with standard descriptive text such as "Trajectory shape is a normal ellipse".

[0147] (3) Description of parameter anomalies

[0148] For slowly changing monitoring parameters (such as temperature, current, vibration amplitude, etc.) during industrial equipment operation, statistical feature extraction and trend analysis are performed within a set time window to identify typical abnormal behaviors, including: increased parameter fluctuations, exceeding reference thresholds, and rapid upward trends. Standardized natural language fault description text is then generated. Let the parameter time series be:

[0149]

[0150] To ensure that anomaly detection is adaptive, a reference model is built for each type of parameter during the equipment commissioning or stable operation phase, including the following benchmark indicators, for reference in the generation of parameter anomalies;

[0151] Historical average:

[0152] Historical standard deviation:

[0153] Dynamic threshold range:

[0154] in, For the first Parameter values ​​for each sampling period, For the number of historical samples, This is the historical mean of the parameter, used to describe the long-term average level. The historical standard deviation of the parameter is used to measure the fluctuation range of the parameter. These are the dynamic lower and upper threshold values ​​for the parameters. This is a relaxation factor used to adjust the threshold range; it is typically set to... The range between these values ​​is used to dynamically determine whether a parameter is abnormal and can adapt to the fluctuation characteristics of different parameters.

[0155] In an optional implementation, the anomaly feature extraction and generation includes:

[0156] a. Volatility anomalies

[0157] Calculate the standard deviation of the sampled data within the current time window:

[0158]

[0159] in, This represents the average value of the parameter within the current time window.

[0160] If the following conditions are met:

[0161]

[0162] This indicates a significant increase in parameter volatility, and a natural language description of the parameter characteristics is generated:

[0163] Output begins: "This parameter fluctuates significantly and shows an unstable trend."

[0164] in, This is a relaxation factor used to adjust the threshold range; it is typically set to... The values ​​between.

[0165] b. Exceeding limits

[0166] For each sampling point Execute threshold judgment:

[0167] or

[0168] If true, then the current parameters are considered to be outside the safe operating range, and a text description of the parameter characteristics is generated:

[0169] Output begins: "The current value of this parameter has exceeded the set threshold range, posing a potential risk."

[0170] c. Abnormal Trend

[0171] To evaluate the trend of parameter changes over time, the system... Perform linear fitting on the point set and calculate the trend slope:

[0172]

[0173] If the following conditions are met:

[0174]

[0175] This indicates that the parameter is showing a rapid upward trend, and a natural language description of the parameter characteristics is generated:

[0176] Output begins: "This parameter has risen rapidly in a short period of time, which may indicate a persistent abnormal trend."

[0177] in, This is a relaxation factor used to adjust the threshold range; it is typically set to... The values ​​between.

[0178] When multiple features simultaneously meet the abnormal conditions, multiple information segments can be automatically combined to form a complete structured fault feature description text, such as: > "The temperature parameter has risen rapidly in the past 10 minutes, with a slope of 2.4℃ / min, which is significantly higher than the historical average rise rate of 1.0℃ / min. The current value of 58.2℃ has exceeded the threshold limit of 55℃, and the volatility has increased significantly."

[0179] (4) Synthesis of fault feature descriptions

[0180] The natural language descriptions of various fault characteristics, such as vibration spectrum, shaft center trajectory, and parameter trends, are merged to generate a unified text description of fault characteristics. As one of the input contents for reasoning:

[0181]

[0182] in, The descriptive text generated for spectrum analysis (e.g., the fundamental frequency of a certain measurement point is 100Hz, there is a 2X harmonic, and an abnormal frequency peak of 1000Hz appears). The result of the axis trajectory analysis (e.g., the axis trajectory is in the shape of an 8). Based on trend and exceedance analysis results (such as rapid temperature rise, with the current value of 58.2℃ exceeding the threshold upper limit of 55℃), the final synthesized As the main input field of the fault knowledge retrieval and reasoning diagnosis module, it is structurally a natural language concatenation result and content-wise a semantic representation of structured features.

[0183] It should be noted that the fault knowledge base retrieval uses textual descriptions of fault characteristics as query conditions to retrieve both structured and unstructured fault knowledge. It supports causal chain completion and professional support in large-scale model inference tasks, including:

[0184] (1) Fault Vector Knowledge Base

[0185] The knowledge base content comes from industrial field data, including but not limited to equipment manuals, maintenance manuals, historical failure case records, expert experience documents and Q&A records;

[0186] All knowledge texts are segmented according to the structure of "equipment type + faulty component + fault characteristics". Let the first segment be... The knowledge segment is as follows:

[0187]

[0188] By embedding model Convert text into a high-dimensional vector:

[0189]

[0190] All embedded vectors are stored in a vector database to build an index.

[0191] (2) Query and similarity retrieval

[0192] Fault characteristic text description (For example, if the axis trajectory is in the shape of an 8, and the vibration spectrum shows a 2X harmonic,) this is also converted into a query vector through an embedding model:

[0193]

[0194] in, A textual description of the fault characteristics. For text embedding models, For text description The resulting query vector;

[0195] Perform vector similarity retrieval and calculate the query vector. With knowledge fragment vector set Similarity:

[0196]

[0197] in, This is a vector representation of a knowledge fragment (document fragment). For vector dot product operation, , Let L2 be the L2 norm (length) of the vector. To query the cosine similarity between a vector and a knowledge fragment vector, the range of values ​​is given. The closer the value is to 1, the higher the correlation.

[0198] Perform a Top-K search based on similarity and select the most relevant results. Knowledge snippet:

[0199]

[0200] in, The most relevant ones retrieved at the end A collection of knowledge fragments For the first A knowledge fragment, To sort by similarity value from highest to lowest, and then take the top... The operation of the bar The number of knowledge fragments returned by the retrieval;

[0201] Finally, a collection of knowledge fragments is returned. The content includes maintenance knowledge, case studies, common fault causes and handling suggestions that are most relevant to the current fault characteristics, and is used to concatenate with the prompts in the inference model.

[0202] Furthermore, the structured fault characteristic description A collection of knowledge fragments related to recall The concatenation into unified semantic prompts serves as input to a large language model, which then performs tasks such as fault type determination, causal chain construction, and maintenance suggestion generation through an inference model; where:

[0203] (1) Construction of input for fault reasoning

[0204] : Generated textual description of fault characteristics (natural language expression);

[0205] Semantic vector retrieval recall A fragment of fault knowledge;

[0206] : A prompt template structure used for reasoning;

[0207] For example, the input context of the inference model is represented as follows:

[0208]

[0209] For example, the cue word concatenation structure include:

[0210] You are an industrial equipment intelligent diagnostics expert with experience in the operation and maintenance of industrial equipment. Based on the fault characteristic descriptions provided below, and in conjunction with the provided "Fault Knowledge Base," please conduct professional fault diagnosis and analysis.

[0211] # Diagnostic requirements

[0212] If the fault characteristics do not match the fault knowledge well, please output "No fault" in the fault type, "Low" in the diagnostic result confidence, and empty values ​​in other fields.

[0213] The description of the fault symptoms must strictly adhere to the fault characteristics; fabricated characteristics are not permitted.

[0214] # Output Format

[0215] Please output a structured fault diagnosis report, including:

[0216] - "Fault Type": Matches the most likely fault name

[0217] - "Fault Manifestation": Describe the intuitive phenomena of the fault in conjunction with its characteristics.

[0218] - "Fault Analysis and Reasoning Process": Provide a detailed explanation of the diagnostic cause, citing the analysis process including parameter trends, spectrum, vibration, and shaft center characteristics, etc., with a minimum of 100 words.

[0219] - "Diagnostic Result Confidence": Values ​​are "High", "Medium", or "Low". The confidence level of the diagnostic result reflects the degree of match between the diagnostic result and the fault type characteristic description. The confidence level should be as rigorous as possible, with "High" indicating a high degree of match, "Medium" indicating a moderate degree of match, and "Low" indicating a low degree of match.

[0220] - "Repair Recommendations": Provide professional inspection or maintenance advice.

[0221] # Fault characteristics:

[0222] {Fault Feature Text Description Q}

[0223] # Fault Knowledge Base:

[0224] {Related Fault Knowledge K} top}

[0225] (2) Inference model function

[0226] The reasoning model is a large language model. Its parameters are This can be a locally deployed model (such as Deepseek-7B, Qwen3-4B):

[0227]

[0228] This function completes the mapping from fault symptoms and background knowledge to diagnostic conclusions.

[0229] (3) Output structure definition

[0230] The model outputs a structured diagnostic report:

[0231]

[0232] Specifically, it is expressed as follows:

[0233] Fault type: ;

[0234] Fault description: Natural language text;

[0235] Reasoning process: A logical chain combining features such as spectrum, trend, and trajectory;

[0236] Diagnostic confidence level: ;

[0237] Recommendations: Case-based maintenance recommendations;

[0238] in, Labels indicating the types of faults identified by the system. A predefined set of fault types, such as typical fault categories like bearing wear, rotor imbalance, and coupling misalignment.

[0239] For example, the fault reasoning diagnosis result output includes:

[0240] Fault type: Coupling misalignment

[0241] Fault symptoms: The bearing temperature of the equipment rose from 20℃ to 53℃, the vibration value reached 75 mm / s (far exceeding the 20 mm / s threshold), the vibration spectrum showed obvious 2X harmonic components, and the shaft center trajectory was in the shape of an "8".

[0242] Fault analysis and reasoning process: Based on sensor characteristic data, the equipment vibration was first observed to be severely excessive (75 mm / s), far exceeding the fault threshold of 20 mm / s, indicating a significant mechanical problem. The vibration spectrum showed a clear 2X harmonic characteristic, a typical feature of coupling misalignment. Secondly, the shaft trajectory exhibited an "8" shape, a hallmark of coupling misalignment in mechanical fault diagnosis. Furthermore, the bearing temperature rapidly increased from 20℃ to 53℃, which, while not exceeding the extreme temperature, combined with the intense vibration, indicated a serious misalignment or imbalance problem in the rotor system.

[0243] Compared with the coupling misalignment item in the fault knowledge base, its typical characteristics are: "prominent 2X frequency in the spectrum", "shaft center trajectory is figure 8", and "increased vibration of the main shaft and motor", which are highly consistent with the current characteristics. Other fault types such as "bearing damage", "shaft bending" and "internal friction" may show some vibration increase or temperature rise, but they do not match in terms of frequency spectrum and trajectory characteristics.

[0244] Therefore, the most likely fault, based on comprehensive analysis, is misalignment of the coupling.

[0245] Confidence level of diagnostic results: High

[0246] Repair suggestions:

[0247] 1. After shutdown, focus on checking the alignment of the shafts on both sides of the coupling, and use a laser alignment instrument to recalibrate the coupling installation.

[0248] 2. Check the coupling body for wear or structural defects such as loose bushings, and replace parts if necessary.

[0249] 3. Inspect for secondary damage caused by misalignment, such as bearing wear and shaft seal deformation, and address them simultaneously.

[0250] 4. After troubleshooting, be sure to retest the vibration and temperature baselines before restarting to ensure the problem is completely resolved.

[0251] 5. It is recommended to strengthen online monitoring and trend analysis of the 2X frequency multiplier during operation, and establish an early warning mechanism for changes in the operating status.

[0252] Reference Figure 4 Other aspects disclosed in the embodiments of the present invention also propose an industrial fault diagnosis system based on natural language fault features and large model knowledge-enhanced reasoning, including:

[0253] The multi-source sensor module is used to collect signals such as vibration, temperature, sound and electrical quantities of the device under test, and output them through a standard industrial interface;

[0254] The data acquisition and analysis unit is electrically connected to the multi-source sensor module and is used to perform signal conditioning, analog-to-digital conversion, filtering and standardization on the acquired signals and generate status analysis results and fault characteristic data.

[0255] It should be noted that the data acquisition and analysis unit consists of a sensor interface module, a signal conversion and processing module, a data buffer module, a status recognition module, and an analysis model module.

[0256] The fault diagnosis reasoning unit is connected to the data acquisition and analysis unit. It is used to transcribe fault feature data into fault feature text, perform vector retrieval on the fault knowledge base and perform large model reasoning diagnosis, and output structured diagnosis results.

[0257] For example, the fault diagnosis reasoning unit includes an embedded AI chip, an edge reasoning module, a local knowledge base cache structure, an embedded hot database, a model reasoning engine, and an interface processing module, for fault feature text generation, fault knowledge base retrieval, and fault reasoning.

[0258] The results display service module communicates with the fault diagnosis reasoning unit to receive and provide diagnostic results, evidence fragments, and original / derived data in a service manner.

[0259] It should be noted that the results display service module includes a page display service module, a data interface service module, and an intelligent agent workflow module. The intelligent agent workflow module is used to trigger knowledge retrieval, question-and-answer linkage, and alarm linkage processes based on the running status or user requests.

[0260] The display unit, connected to the fault diagnosis reasoning unit and / or result display service module, is used to display spectrum diagrams, shaft center trajectory, parameter trends, alarm information and structured diagnostic results in a graphical interface, and supports interactive queries.

[0261] The external communication and data interface module connects to the fault diagnosis reasoning unit and the result display service module to realize data interaction and remote synchronization with the host computer or PLC system.

[0262] For example, the external communication and data interface module includes a USB interface, an RS485 serial port, an HDMI interface, an Ethernet interface, and an audio output interface;

[0263] The power management module is used to supply power to the above-mentioned units and modules and provide overvoltage, undervoltage and overcurrent protection;

[0264] The multi-source sensor module transmits processed feature data to the fault diagnosis reasoning unit via the data acquisition and analysis unit. The fault diagnosis reasoning unit then provides the structured diagnostic results to the result display service module and the display unit for presentation, and outputs them to the outside world via the external communication and data interface module.

[0265] The system also includes one or more processors and memory.

[0266] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning as described in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0267] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the industrial fault diagnosis method based on natural language fault features and large model knowledge-enhanced reasoning of the foregoing embodiments, particularly... Figure 1 The flowchart of the method is shown.

[0268] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0269] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0270] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if required, the program can be implemented in assembly or machine language.

[0271] In any case, the language can be either compiled or interpreted.

[0272] Furthermore, for this purpose, the program can run on programmed application-specific integrated circuits.

[0273] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0274] Furthermore, the method can be implemented in any suitable computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0275] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0276] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0277] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0278] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An industrial fault diagnosis method based on natural language fault features and large model knowledge enhanced reasoning, characterized in that, Comprise: In the device running phase, obtain multi-source monitoring and state analysis results from the device under test, which at least include vibration spectrum features, shaft center trajectory features and running parameter trend features; The acquisition of multi-source monitoring and state analysis results from the device under test comprises: Performing fast Fourier transform on the vibration original signal, extracting representative spectral peaks based on a preset peak prominence threshold, and merging similar peaks according to frequency resolution to determine the fundamental frequency and its multiples; Collecting X and Y direction displacement to form a shaft center trajectory point set, performing centroid translation and main direction alignment on the trajectory, calculating curvature disturbance and distance disturbance and detecting the number of trajectory intersection points, and determining the trajectory as 8-shaped, flat ellipse, banana-shaped or normal ellipse; Calculating the mean, standard deviation and trend slope of slow-changing parameters of temperature, current and vibration amplitude in a sliding time window, and combining with dynamic threshold interval to determine over-limit, volatility anomaly and trend anomaly to form structured state analysis results; Converting the features into structured natural language fault feature text: Merging the natural language description results of vibration spectrum, shaft center trajectory and running parameter trend features to generate a unified fault feature text description as one of the reasoning input contents; Vectorizing the fault feature text, performing similarity retrieval in the pre-built industrial fault knowledge base, and recalling knowledge fragments related to device type, component and working condition: Embedding model encoding the fault feature text into a query vector; Segmenting and embedding the unstructured knowledge of maintenance manual, historical cases and expert experience according to device type, fault component and fault feature, and establishing vector index; Performing Top-K retrieval based on cosine similarity, filtering according to device model, working condition label and time stamp, and returning the most relevant knowledge fragments and their source identifiers to the current scene; The industrial fault knowledge base includes structured data and unstructured documents, wherein: The structured data includes parameter threshold table, component hierarchy and common fault cause table; The unstructured documents include maintenance manual, historical fault cases and expert experience entries; Segmenting the unstructured documents by paragraphs and adding identifiers, and embedding them into vectors by embedding model to establish vector index to support Top-K retrieval and filtering based on device model, working condition label and time stamp; Based on the fault feature text and recalled knowledge fragments, build reasoning prompts and input them into large language models for knowledge-enhanced reasoning to generate diagnosis results including fault type, affected component, diagnosis basis and disposal suggestion: Build reasoning prompts according to role, task, constraint and format templates; Concatenate the fault feature text and Top-K knowledge fragments into a large language model, and output fault type, fault performance, fault analysis reasoning process, diagnosis result confidence and maintenance suggestion according to structured fields; When the matching degree of the fault feature text and Top-K knowledge fragments is lower than the threshold, the fault type output is no fault, the diagnosis result confidence output is low, and the remaining fields are left blank.

2. The industrial fault diagnosis method based on natural language fault features and large model knowledge enhanced reasoning according to claim 1, characterized in that, The generation of the fault feature text comprises: The spectrum feature extraction extracts the main peak frequency, the amplitude ratio of the frequency multiplication / sub-band interval, and generates a sentence according to a preset word template; The axis trajectory feature obtains the trajectory shape, and generates a sentence according to a preset word template; The parameter trend feature calculates the mean, volatility and trend slope, and maps the quantization results to a grade label to synthesize a multi-sentence structured text description.

3. The industrial fault diagnosis method based on natural language fault features and large model knowledge enhanced reasoning according to claim 1, characterized in that, The construction of the reasoning prompt word adopts a four-element template of role, task, constraint and format, and the four-element template at least includes: The role is set as an industrial fault diagnosis expert; The task is to combine the fault feature text and the knowledge fragment to perform causal analysis and give executable disposal suggestions; The constraint includes unit unification, component name standardization, and conclusion traceable to specific evidence items; The format requires output in a structured field.

4. An industrial fault diagnosis system based on natural language fault features and large model knowledge enhanced reasoning, characterized in that, It includes: A multi-source sensor module for collecting vibration, temperature, sound, and electrical signals of the measured equipment and outputting through a standard industrial interface; A data acquisition and analysis unit electrically connected with the multi-source sensor module for signal conditioning, analog-to-digital conversion, filtering and standardization processing of the collected signals and generating state analysis results and fault feature data; A fault diagnosis reasoning unit in communication connection with the data acquisition and analysis unit for converting the fault feature data into fault feature text, performing vector retrieval on the fault knowledge base and large model reasoning diagnosis, and outputting structured diagnosis results; A result display service module in communication connection with the fault diagnosis reasoning unit for receiving and providing diagnosis results, evidence fragments and original / derived data in a service manner; A display unit for displaying spectrum graphs, axis trajectories, parameter trends, alarm information and the structured diagnosis results in a graphical interface and supporting interactive queries; An external communication and data interface module for realizing data interaction and remote synchronization with a host computer or a PLC system; A power management module for supplying power to the above units and modules and providing overvoltage, undervoltage and overcurrent protection.

5. The industrial fault diagnosis system based on natural language fault features and large model knowledge enhanced reasoning according to claim 4, characterized in that, The system further includes one or more processors; A memory storing executable instructions, which when executed by the one or more processors, cause the one or more processors to perform operations including the flow of the industrial fault diagnosis method based on natural language fault feature and large model knowledge enhanced reasoning according to any one of claims 1-3.

6. A computer readable medium storing software, characterized in that: The software includes instructions executable by one or more computers, which through such execution cause the one or more computers to perform operations including the flow of the industrial fault diagnosis method based on natural language fault feature and large model knowledge enhanced reasoning according to any one of claims 1-3.

Images