Rotating machinery continuous diagnosis method based on future state prediction and world model

Abstract

The application discloses a rotating machinery continuous diagnosis method based on future state prediction and a world large model, and the method comprises the following steps: acquiring a multi-source data sequence of the rotating machinery and driving a digital twin model; generating a first future state prediction sequence of the rotating machinery in a future period of time based on the digital twin model; jointly coding the multi-source data sequence and the first future state prediction sequence into a prompt vector, inputting the prompt vector into the world large model, outputting a diagnosis result and a decision suggestion, and outputting a second future state prediction sequence of the rotating machinery in the same future period of time; when a new failure mode is identified, starting a continuous learning mechanism, introducing a low-rank matrix, and using a total loss function comprising a task loss, a future information consistency loss and a knowledge distillation loss to continuously learn and update parameters of the world large model. The application provides a rotating machinery fault diagnosis method capable of realizing prospective prediction, continuous knowledge evolution and decision explanation.

Description

Technical Field

[0001] This invention belongs to the field of rotating machinery fault diagnosis technology, specifically relating to a method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model. Background Technology

[0002] Rotating machinery (such as gas turbines, wind turbines, and aircraft engines) is core equipment in energy, transportation, and manufacturing industries, and its health status directly affects production safety and economic efficiency. Traditional fault diagnosis methods, primarily based on the analysis of historical data and current conditions, have the following limitations:

[0003] (1) Passivity: It is mostly diagnosis after the fault occurs or simple early warning based on fixed thresholds, lacking forward-looking prediction of the fault evolution trend.

[0004] (2) Staticity: Once the diagnostic model is trained, its knowledge system is fixed, making it difficult to adapt to the natural decline of equipment performance and new, unknown failure modes, and it is prone to catastrophic amnesia.

[0005] (3) Isolation: It relies on data from specific equipment or operating conditions, has poor generalization ability, and the diagnostic results are mostly simple classification labels, lacking in-depth interpretation of fault mechanisms, development trends and maintenance decisions.

[0006] In recent years, pre-trained large models have demonstrated powerful knowledge aggregation and logical reasoning capabilities in fields such as natural language processing. Meanwhile, digital twin technology has made it possible to construct virtual mappings of physical devices. However, how to combine these technologies with fault diagnosis and address the issues of continuous learning and forward-looking prediction remains a challenging problem that current technologies have not yet adequately solved. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model. This method combines digital twins, large language models, and continuous learning technologies to achieve forward-looking prediction, continuous knowledge evolution, and decision interpretation in the diagnosis of rotating machinery faults.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] The first aspect of the present invention provides a method for continuous diagnosis of rotating machinery based on future state prediction and a world-large model, the method comprising:

[0010] Acquire multi-source data sequences of rotating machinery and use these multi-source data sequences to drive the digital twin model of the rotating machinery, so as to achieve state synchronization between the digital twin model and the physical entity;

[0011] Based on the digital twin model, a first future state prediction sequence of rotating machinery is generated over a period of time.

[0012] The multi-source data sequence and the first future state prediction sequence are jointly encoded into a cue vector, which is then input into a pre-trained world big model to output diagnostic results and decision suggestions, and output the second future state prediction sequence of the rotating machinery within the same future time period.

[0013] When a new fault mode is identified, a continuous learning mechanism is initiated. This mechanism introduces a low-rank matrix into the world-wide model through low-rank fine-tuning and uses a total loss function, including task loss, future information consistency loss, and knowledge distillation loss, to continuously learn and update the parameters of the world-wide model. Among them, task loss and knowledge distillation loss are used to ensure that the world-wide model correctly classifies new fault data and old fault data, respectively. Future information consistency loss is used to ensure that the updated world-wide model's future state predictions for new fault data are consistent with the physical predictions of the digital twin model by constraining the second future state prediction sequence and the first future state prediction sequence.

[0014] Furthermore, the multi-source data sequence includes multi-source data vectors at multiple time steps, where each time step's multi-source data vector includes vibration, sound, and temperature data.

[0015] Furthermore, the future state prediction sequence includes performance degradation indicators and key frequency component amplitudes.

[0016] Furthermore, through the open set detection function New failure modes have been identified, as follows:

[0017] ;

[0018] In the formula, This represents an open set detection function; This represents a multi-source data sequence for rotating machinery; This represents the first future state prediction sequence; Indicates the confidence threshold;

[0019] When detected Greater than the confidence threshold When this occurs, it is considered that a new fault mode has been identified.

[0020] Furthermore, open set detection function Implemented using any of the following methods:

[0021] (1) Train the autoencoder to learn to reconstruct known fault modes; after training, input the new multi-source data sequence and the corresponding first future state prediction sequence into the autoencoder, calculate the reconstruction error, and if the reconstruction error is greater than the confidence threshold, If so, it is considered that a new fault mode has been identified;

[0022] (2) Based on the cue vectors of each fault category, calculate a fault prototype for each fault category; obtain the cue vectors of the new multi-source data sequence and the corresponding first future state prediction sequence, calculate the minimum distance with all fault prototypes, and if the minimum distance is greater than the confidence threshold, If so, it is considered that a new fault mode has been identified;

[0023] (3) When outputting diagnostic results and decision recommendations, the world-class model also outputs a confidence level. If the difference between 1 and the confidence level is greater than the confidence level threshold, the model will be considered as having a confidence level. If so, it is considered that a new fault mode has been identified.

[0024] Furthermore, the total loss function is as follows:

[0025] ;

[0026] In the formula, Indicates the total loss; This represents task loss, used to ensure that the world-class model correctly classifies new faulty data. This represents the knowledge distillation loss, used to ensure that the world-large model correctly classifies old faulty data. This represents the loss of future information consistency, used to constrain the updated world-large model's prediction of the future state of new fault data to be consistent with the physical prediction of the digital twin model; and This represents the weighting factor.

[0027] Furthermore, the future loss of information consistency is as follows:

[0028] ;

[0029] In the formula, This indicates a loss of future information consistency. This indicates new fault data. Representing a digital twin model, This represents the physical prediction of the digital twin model, i.e., the first future state prediction sequence; Indicates encoding, Represents the prompt vector; Represents the parameters of the world-large model; This represents the incremental update of the parameters of the world-large model, determined by the introduced low-rank matrix; This represents the updated world model. This represents the output of the updated world model, including the updated world model's prediction of the future state of the new fault data, i.e., the second future state prediction sequence.

[0030] Furthermore, the knowledge distillation loss is as follows:

[0031] ;

[0032] In the formula, This represents the loss from knowledge distillation. This indicates old fault data; Represents the old world grand model, This represents the updated world model; represents the Softmax function, and KL represents the Kullback-Leibler divergence.

[0033] According to a second aspect of the present invention, a continuous diagnostic system for rotating machinery based on future state prediction and a world-scale model is provided, applying the continuous diagnostic method for rotating machinery based on future state prediction and a world-scale model as described in any one of the first aspects. The system comprises:

[0034] The data acquisition and preprocessing module is used to acquire and preprocess multi-source data from rotating machinery to generate multi-source data sequences.

[0035] The digital twin and future prediction module is used to drive the digital twin model of rotating machinery using multi-source data sequences, so as to achieve state synchronization between the digital twin model and the physical entity, and generate the first future state prediction sequence of the rotating machinery for a period of time in the future based on the digital twin model.

[0036] The world-class model diagnostic reasoning module is used to encode the multi-source data sequence and the first future state prediction sequence into a prompt vector, which is then input into the pre-trained world-class model to output diagnostic results and decision suggestions, and output the second future state prediction sequence of the rotating machinery within the same future time period.

[0037] The continuous learning and model update module is used to initiate a continuous learning mechanism when a new fault mode is identified. This mechanism introduces a low-rank matrix into the world-wide model through low-rank fine-tuning and uses a total loss function, including task loss, future information consistency loss, and knowledge distillation loss, to continuously learn and update the parameters of the world-wide model. Among them, task loss and knowledge distillation loss are used to ensure that the world-wide model correctly classifies new fault data and old fault data, respectively. Future information consistency loss is used to ensure that the updated world-wide model's future state predictions for new fault data are consistent with the physical predictions of the digital twin model by constraining the second future state prediction sequence and the first future state prediction sequence.

[0038] According to a third aspect of the present invention, a computer device is provided, comprising: a processor and a memory, the memory storing a program or instructions executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method for continuous diagnosis of rotating machinery based on future state prediction and a world-large model as described in any one of the first aspects.

[0039] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a program or instructions stored thereon, which, when executed by a processor, implement the steps of the method for continuous diagnosis of rotating machinery based on future state prediction and a world-large model as described in any one of the first aspects.

[0040] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0041] (1) Foresight and initiative: By introducing future state information as guidance, fault diagnosis is moved from post-event and in-event to pre-event, realizing true predictive maintenance and providing a valuable lead time for scheduling maintenance.

[0042] (2) Continuous evolution capability: Through continuous learning technologies such as low-rank fine-tuning guided by future information, the system can continuously learn new fault modes and adapt to equipment degradation without forgetting old knowledge, thus becoming an expert system that never falls behind.

[0043] (3) Strong credibility and solvability of decision-making: By utilizing the powerful knowledge base and natural language generation capabilities of the world big model, it can not only provide diagnostic conclusions, but also provide detailed reasoning processes, fault mechanism analysis and maintenance suggestions, which greatly improves the credibility and engineering applicability of the results.

[0044] (4) High generalization and knowledge reusability: The pre-trained world large model itself gathers general knowledge across devices and working conditions, which makes it more capable of initial generalization when facing new models of equipment or complex working conditions.

[0045] In summary, this invention combines digital twins, large language models, and continuous learning technologies to achieve forward-looking prediction, continuous knowledge evolution, and decision interpretation in rotating machinery fault diagnosis. Attached Figure Description

[0046] Figure 1 This is an overall architecture diagram of a continuous diagnostic system for rotating machinery based on future state prediction and a world-large model, provided by an embodiment of the present invention.

[0047] Figure 2 A schematic diagram of the overall process of a continuous diagnostic method for rotating machinery based on future state prediction and a world-large model, provided in an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of a continuous learning mechanism guided by future state information provided in an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of the hardware structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0051] Obviously, the accompanying drawings described below are merely some examples or embodiments of the present invention. Those skilled in the art can apply the present invention to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, modifications to design, manufacturing, or production based on the technical content disclosed in this invention are merely conventional technical means and should not be construed as insufficient disclosure of the present invention.

[0052] In this invention, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this invention may be combined with other embodiments without conflict.

[0053] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "a," "an," "an," "the," and similar words used in this invention do not indicate quantity limitation and may indicate singular or plural. The terms "comprising," "including," "having," and any variations thereof used in this invention are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms "connected," "linked," "coupled," and similar words used in this invention are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "A plurality" used in this invention refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships may exist; for example, "A and / or B" can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship. The terms "first," "second," and "third" used in this invention are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0054] This invention provides a method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model. This method combines digital twins, large language models, and continuous learning techniques, and is a method for diagnosing rotating machinery faults that can achieve forward-looking prediction, continuous knowledge evolution, and decision interpretation.

[0055] like Figure 1 and Figure 2 As shown, the continuous diagnostic method for rotating machinery based on future state prediction and a world-scale model of the present invention includes the following steps:

[0056] Step S1: Multi-source data fusion and digital twin synchronization.

[0057] Real-time multi-source data of rotating machinery is collected and preprocessed to construct and generate multi-source data sequences of rotating machinery. :

[0058] ;

[0059] in, Indicates at time step The multi-sensor reading vectors include vibration, sound, temperature, and other sensors. Indicates the rotating machinery at time step The multi-source data sequence includes a total of A multi-source data vector at each time step.

[0060] Then, this data is used to drive the digital twin model of the physical device. This synchronizes the digital twin model with the physical entity, making its state synchronized. The digital twin model not only replicates the equipment's geometry and material properties but also embeds the physical laws governing its operation (such as dynamics, vibration, and wear evolution). It can use built-in component loss mechanism formulas (such as fatigue wear models and performance degradation equations) combined with real-time operating data (such as load and temperature) to predict future performance degradation trends and output indicators such as effective vibration values, peak factor, and predicted remaining service life. Furthermore, the model replicates the dynamic characteristics of rotating machinery (such as gear meshing laws and bearing rotation principles), and by simulating vibration spectrum changes under future operating conditions, it directly outputs key frequency parameters such as rotational frequency and its harmonic amplitudes, fault characteristic frequency amplitudes, and sideband amplitudes. This is existing technology and will not be elaborated further here.

[0061] Step S2: Prediction and guidance of future state information.

[0062] Based on digital twin model Generate the future Predicting the state sequence within a time period, i.e. generating the rotating machinery in the future First future state prediction sequence within time:

[0063] ;

[0064] in, This information predicts the future state, including performance degradation metrics and key frequency component amplitudes. This future state information is used for early risk identification and model update guidance in subsequent steps.

[0065] For example, performance degradation indicators can include vibration RMS value, peak factor, kurtosis, envelope spectrum entropy, performance degradation index, and remaining service life prediction value; key frequency component amplitudes can include rotational frequency and its harmonic amplitude, bearing fault characteristic frequency amplitude, gear meshing frequency and its sideband amplitude, blade passing frequency amplitude, resonance frequency component amplitude, and high-frequency resonance demodulation band energy.

[0066] Step S3: Fault diagnosis and reasoning of the world large model.

[0067] Real-time data Information about future states Common encoding into a cue vector And input it into the pre-trained world-class model. The world-class model is based on its parameters. Perform reasoning and output diagnostic results. Decision recommendations Simultaneously, it outputs a second future state prediction sequence for the rotating machinery within the same future time period. :

[0068] ;

[0069] in, This indicates an encoding operation.

[0070] Example scenario: Monitoring of wind turbine gearbox.

[0071] enter:

[0072] Real-time data The vibration signal showed an increase in the amplitude of higher harmonics related to the gear meshing frequency, and the cabin temperature showed a slight upward trend.

[0073] Future state information Digital twin models predict the next 24 hours ( Within this range, the harmonic amplitude will continue to increase by 50%, and the sideband energy will increase significantly, while the bearing temperature will exceed the normal threshold.

[0074] Output diagnostic results Decision recommendations :

[0075] (1) Diagnostic results

[0076] Diagnostic conclusion: The intermediate shaft gear of the gearbox has early local pitting corrosion, which has already triggered the precursor to bearing overheating.

[0077] Reasoning and Severity Assessment:

[0078] ① Real-time state correlation: High-order harmonics of the meshing frequency in the current vibration spectrum are typical characteristics of gear surface fatigue. Increased temperature indicates that additional frictional losses have begun to occur;

[0079] ② Future trend prediction: The rapid increase in amplitude and the appearance of sidebands in the digital twin prediction indicate that the pitting area is expanding and may develop into a moderate fault within the next 24-48 hours, leading to a sharp increase in vibration intensity;

[0080] ③ Knowledge Base Verification: Based on historical maintenance records and physical models, this type of gearbox, under a similar vibration characteristic evolution path, has an average risk of tooth breakage after 72 hours. The current risk level is assessed as "high".

[0081] (2) Decision Recommendations

[0082] ① Actions to be performed immediately:

[0083] Confirmatory check: Immediately check the color of the gearbox oil remotely using a high-definition camera. If metal fragments are found, the diagnosis can be further confirmed.

[0084] Operational adjustments: It is recommended to reduce the power output of the wind turbine to 80% of its rated power within the next 12 hours to slow down the rate of failure evolution.

[0085] ② Proactive maintenance plan:

[0086] Recommended repair window: The optimal repair window is within the next 36 hours. At this time, the fault is identified, but secondary damage to adjacent components (such as bearings or shafts) has not yet occurred.

[0087] Spare parts and personnel: The system has automatically generated a work order and notified the warehouse to prepare spare gear sets and bearings. It is also recommended to arrange for the maintenance team to be on standby 24 hours later.

[0088] ③ Long-term monitoring strategy update:

[0089] Model Update: The sequence data from this diagnosis has been labeled as early samples of the new failure mode. After maintenance is completed, the continuous learning module will use this data to perform low-rank fine-tuning on the world's largest model to enhance its ability to identify similar early failures.

[0090] Monitoring parameter adjustment: The monitoring thresholds for vibration harmonic amplitude and bearing temperature will be temporarily lowered by 20%, and a special inspection will be initiated for other units with the same drive chain as this fan.

[0091] (3) Second future state prediction sequence

[0092] The world-wide model, based on its integrated physical knowledge and historical failure cases, provides insights into the same future time period (…). The equipment status within a 24-hour period was predicted twice, and the following predictive judgment was output:

[0093] Based on the current upward trend of vibration harmonics and temperature anomalies in the gearbox, combined with the gearbox dynamics model and similar fault evolution patterns, the following predictions are made for the next 24 hours:

[0094] The vibration harmonic amplitude will increase by about 45% to 55% from the current level, approaching the critical alarm value;

[0095] The sideband energy will increase by 60% to 80%, indicating that the fault is spreading from local pitting to distributed damage.

[0096] The bearing temperature will exceed the normal upper limit for the first time within 12 hours and continue to rise to the warning level within 24 hours. The overall performance degradation index will accelerate from the current 0.12 to 0.28, entering the "medium risk" range.

[0097] This sequence not only reproduces the physical prediction trends of digital twins, but also introduces semantic inferences of the "acceleration effect" and "cross-component impact" of fault propagation, forming a status report with interpretability and early warning depth.

[0098] As a result, the large model uses its internal knowledge to correlate the current state with future trends, generating a natural language report that includes predictive judgments.

[0099] It should be noted that the world-wide model of this invention differs from existing language-wide models. Existing language-wide models are all trained on a foundational transformer architecture, while the world-wide model uses a world model as its foundation. The world model can learn the internal patterns of signals themselves, elevating various available input signals into a complete three-dimensional world, interacting with data from the digital twin model, and continuously iterating and updating its understanding of the world as new information emerges. In this invention, the world-wide model utilizes its internal knowledge to correlate the current state with future trends, generating natural language reports containing predictive judgments.

[0100] Step S4: Continuous learning and model updates based on future information.

[0101] When the system passes the open set detection function When a new failure mode is identified, i.e. (in (As the confidence threshold), the continuous learning mechanism is initiated.

[0102] Open set detection function It is the "trigger" of the entire continuous learning system, and its importance is self-evident. Open set detection acknowledges the existence of "unknown unknowns." The model can not only correctly classify known categories (such as normal, bearing failure, gear wear), but also identify inputs that do not belong to any known category, that is, judge them as "unknown" or "new faults".

[0103] Open set detection function Used to determine in real time whether the current system state (combined with its future predictions) deviates from all known normal or fault modes; if an unknown state is detected and the confidence level exceeds a threshold. This signifies the emergence of new knowledge to be learned, triggering subsequent low-rank fine-tuning and other continuous learning processes.

[0104] This function is a computational process that outputs a confidence score representing "unknown" or "novelty." A higher score indicates a greater likelihood that the current data represents a "new fault." (Open set detection function) This can be achieved in the following ways:

[0105] (1) Method based on reconstruction error

[0106] Train an autoencoder specifically to learn how to perfectly reconstruct data from a "known" state. For known states, the reconstruction error will be small. When a new data sequence is input... and the first future state prediction sequence Then, calculate its reconstruction error:

[0107] ;

[0108] in, Indicates encoding operation. This indicates a decoding operation.

[0109] If the error is greater than the threshold This indicates that the model cannot understand the current state well and is therefore classified as "unknown".

[0110] (2) Distance-based methods

[0111] In the feature space of the model (i.e.) In the output, a prototype is calculated for each known category. For example, class centers can be obtained by clustering the cue vectors of each fault category. Alternatively, the mean of the cue vectors for each fault category can be calculated as the fault prototype for each category.

[0112] Specifically, calculate the cue vector of the new multi-source data sequence and the corresponding first future state prediction sequence. With all known category prototypes Minimum or average distance:

[0113] ;

[0114] In the formula, Indicates the first Fault prototypes for various fault categories.

[0115] If this minimum distance is very large, it means that the current state is far away from all known states, and is therefore classified as "unknown".

[0116] (3) Method based on model confidence

[0117] With the help of the world model The output probability or confidence score enables open set detection. A well-calibrated model will output high confidence for known samples, but hesitate for unknown samples (low confidence score with a uniform distribution). If the difference between 1 and this confidence score is greater than the confidence threshold... If so, it is considered that a new fault mode has been identified.

[0118] Not just looking at the current state It also takes into account predictions of future states. It intelligently determines whether a new situation outside the system's knowledge base has been encountered by analyzing the combined characteristics of the current state and future trends. Once confirmed, it sends a signal to initiate a continuous learning mechanism, allowing the world-wide model to incorporate this new knowledge, thereby achieving continuous learning for the system.

[0119] like Figure 3 As shown, the core update mechanism is as follows:

[0120] Step S41: Low-rank fine-tuning.

[0121] This invention does not directly update all parameters of the large model. Instead, a low-rank matrix BA is introduced (where... , , and rank The weights are updated indirectly using [a certain method / mechanism]. The updated forward propagation process is as follows:

[0122] ;

[0123] in, For the original frozen pre-trained weights, This represents a low-rank increment. This significantly reduces the number of parameters to be trained and improves learning efficiency.

[0124] Step S42: Loss function guided by future information.

[0125] The optimization objectives of the entire continuous learning process consist of three parts:

[0126] ;

[0127] In the formula, Indicates the total loss; This represents task loss, used to ensure that the world-class model correctly classifies new faulty data. This represents the knowledge distillation loss, used to ensure that the world-large model correctly classifies old faulty data. This represents the loss of future information consistency, used to constrain the updated world-large model's prediction of the future state of new fault data to be consistent with the physical prediction of the digital twin model; and This represents the weighting factor.

[0128] This represents the loss of future information consistency, which is the key to this invention. It forces the model to update the new data after each update. Future state prediction With digital twin model Physical predictions Maintaining consistency, thus using physical laws as strong constraints to guide model updates:

[0129] ;

[0130] In the formula, This indicates a loss of future information consistency. This indicates new fault data. Representing a digital twin model, This represents the physical prediction of the digital twin model, i.e., the first future state prediction sequence; Indicates encoding, Represents the prompt vector; Represents the parameters of the world-large model; This represents the incremental update of the parameters of the world-large model, determined by the introduced low-rank matrix; This represents the updated world model. This represents the output of the updated world model, including the updated world model's prediction of the future state of the new fault data, i.e., the second future state prediction sequence.

[0131] This represents the knowledge distillation loss, achieved by comparing the old and new models on historical task data. The output is designed to prevent catastrophic forgetting.

[0132] ;

[0133] in, is the Softmax function, and KL is the Kullback-Leibler divergence.

[0134] Through optimization This is used to update the low-rank matrix BA, enabling the model to evolve safely, efficiently, and continuously under the guidance of physical laws.

[0135] In summary, this invention provides a method for continuous fault diagnosis of rotating machinery based on future state prediction and a large-scale model. The method includes: collecting real-time operating data of the rotating machinery; predicting future state information of the equipment based on the data; inputting the real-time data and the future state information into a pre-trained large-scale model to obtain fault diagnosis and predictive judgment results; and when a new fault mode or performance drift is detected, using the future state information as guidance to perform efficient incremental updates of the parameters of the large-scale model. Specifically, the incremental updates of the large-scale model are performed by optimizing a total loss function. This is achieved by including at least one task loss in the total loss function. A loss of consistency with future information Future loss of information consistency This is used to ensure that the predictions of the future states of the input data after the model update remain consistent with the predictions of the physical model; incremental updates are implemented using low-rank fine-tuning techniques, specifically by adjusting the weight matrix of the large model. Introduce a low-rank increment Update, including the rank satisfy .

[0136] Specifically, future information consistency loss The calculation method is as follows: after the model is updated, the new data is processed. and its future information Predicted output With digital twin model right Prediction results The mean square error or cosine distance between them.

[0137] This invention also provides a continuous diagnostic system for rotating machinery based on future state prediction and a world-scale model. The method described in the above embodiments, namely the continuous diagnostic system for rotating machinery based on future state prediction and a world-scale model, is applied as follows: Figure 1 As shown, the system includes:

[0138] The data acquisition and preprocessing module is used to acquire and preprocess multi-source data from rotating machinery to generate multi-source data sequences.

[0139] The digital twin and future prediction module is used to drive the digital twin model of rotating machinery using multi-source data sequences, so as to achieve state synchronization between the digital twin model and the physical entity, and generate the first future state prediction sequence of the rotating machinery for a period of time in the future based on the digital twin model.

[0140] The world-class model diagnostic reasoning module is used to encode the multi-source data sequence and the first future state prediction sequence into a prompt vector, which is then input into the pre-trained world-class model to output diagnostic results and decision suggestions, and output the second future state prediction sequence of the rotating machinery within the same future time period.

[0141] The continuous learning and model update module is used to initiate a continuous learning mechanism when a new fault mode is identified. This mechanism introduces a low-rank matrix into the world-wide model through low-rank fine-tuning and uses a total loss function, including task loss, future information consistency loss, and knowledge distillation loss, to continuously learn and update the parameters of the world-wide model. Among them, task loss and knowledge distillation loss are used to ensure that the world-wide model correctly classifies new fault data and old fault data, respectively. Future information consistency loss is used to ensure that the updated world-wide model's future state predictions for new fault data are consistent with the physical predictions of the digital twin model by constraining the second future state prediction sequence and the first future state prediction sequence.

[0142] Thus, this invention provides an intelligent fault diagnosis method and system for rotating machinery that combines digital twins, large language models, and continuous learning technologies, enabling forward-looking prediction, continuous knowledge evolution, and decision interpretation.

[0143] It should be noted that the steps shown in the above process or in the flowchart of the accompanying figures can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0144] In addition, combined Figure 1 and Figure 2 The continuous diagnostic method for rotating machinery based on future state prediction and a world-large model, as described in this embodiment of the invention, can be implemented by a computer device. Figure 4 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Figure 4 As shown, the device may include a processor 301 and a memory 302 storing computer program instructions.

[0145] Specifically, the processor 301 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0146] Memory 302 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to a data processing device. In a particular embodiment, memory 302 is non-volatile memory. In a particular embodiment, memory 302 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0147] The memory 302 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 301.

[0148] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the rotating machinery continuous diagnosis methods based on future state prediction and world model in the above embodiments.

[0149] In some embodiments, the computer device may further include a communication interface 303 and a bus 300. For example, Figure 4 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 300 and complete communication with each other.

[0150] The communication interface 303 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 303 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0151] Bus 300 includes hardware, software, or both, that couples components of a computer device together. Bus 300 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 300 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a MicroChannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 300 may include one or more buses. Although specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0152] This computer device can execute the continuous diagnostic method for rotating machinery based on future state prediction and a world-large model, as described in this embodiment of the invention, thereby achieving a combination of... Figure 2 This describes a method for continuous diagnostics of rotating machinery based on future state predictions and a world-wide model.

[0153] Furthermore, in conjunction with the continuous diagnostic method for rotating machinery based on future state prediction and a world-scale model in the above embodiments, this invention can be implemented using a computer-readable storage medium. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the continuous diagnostic methods for rotating machinery based on future state prediction and a world-scale model in the above embodiments.

[0154] It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. In addition, depending on the implementation needs, the various steps / components described in this invention can be broken down into more steps / components, or two or more steps / components or parts of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0155] It will be readily understood by those skilled in the art that the above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model, characterized in that, The method includes: Acquire multi-source data sequences of rotating machinery and use these multi-source data sequences to drive the digital twin model of the rotating machinery, so as to achieve state synchronization between the digital twin model and the physical entity; Based on the digital twin model, a first future state prediction sequence of rotating machinery is generated over a period of time. The multi-source data sequence and the first future state prediction sequence are jointly encoded into a cue vector, which is then input into a pre-trained world big model to output diagnostic results and decision suggestions, and output the second future state prediction sequence of the rotating machinery within the same future time period. When a new fault mode is identified, a continuous learning mechanism is initiated. This mechanism introduces a low-rank matrix into the world-wide model through low-rank fine-tuning and uses a total loss function, including task loss, future information consistency loss, and knowledge distillation loss, to continuously learn and update the parameters of the world-wide model. Among them, task loss and knowledge distillation loss are used to ensure that the world-wide model correctly classifies new fault data and old fault data, respectively, and future information consistency loss is used to ensure that the updated world-wide model's future state predictions for new fault data are consistent with the physical predictions of the digital twin model by constraining the second future state prediction sequence and the first future state prediction sequence. The total loss function is as follows: ； In the formula, Indicates the total loss; This represents task loss, used to ensure that the world-class model correctly classifies new faulty data. This represents the knowledge distillation loss, used to ensure that the world-large model correctly classifies old faulty data. This represents the loss of future information consistency, used to constrain the updated world-large model's prediction of the future state of new fault data to be consistent with the physical prediction of the digital twin model; and Indicates the weighting factor; The loss of future information consistency is as follows: ； In the formula, This indicates a loss of future information consistency. This indicates new fault data. Representing a digital twin model, This represents the physical prediction of the digital twin model, i.e., the first future state prediction sequence; Indicates encoding, Represents the prompt vector; Represents the parameters of the world-large model; This represents the incremental update of the parameters of the world-large model, determined by the introduced low-rank matrix; This represents the updated world model. This represents the output of the updated world model, including the updated world model's prediction of the future state of the new fault data, i.e., the second future state prediction sequence.

2. The method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model as described in claim 1, characterized in that, The multi-source data sequence consists of multi-source data vectors at multiple time steps, where each time step's multi-source data vector includes vibration, sound, and temperature data.

3. The method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model as described in claim 1, characterized in that, The future state prediction sequence includes performance degradation indicators and key frequency component amplitudes.

4. The method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model as described in claim 1, characterized in that, Using open set detection function New failure modes have been identified, as follows: ； In the formula, This represents an open set detection function; This represents a multi-source data sequence for rotating machinery; This represents the first future state prediction sequence; Indicates the confidence threshold; When detected Greater than the confidence threshold When this occurs, it is considered that a new fault mode has been identified; Among them, open set detection function Implemented using any of the following methods: (1) Train the autoencoder to learn to reconstruct known fault modes; after training, input the new multi-source data sequence and the corresponding first future state prediction sequence into the autoencoder, calculate the reconstruction error, and if the reconstruction error is greater than the confidence threshold, If so, it is considered that a new fault mode has been identified; (2) Based on the cue vectors of each fault category, calculate a fault prototype for each fault category; obtain the cue vectors of the new multi-source data sequence and the corresponding first future state prediction sequence, calculate the minimum distance with all fault prototypes, and if the minimum distance is greater than the confidence threshold, If so, it is considered that a new fault mode has been identified; (3) When outputting diagnostic results and decision recommendations, the world-class model also outputs a confidence level. If the difference between 1 and the confidence level is greater than the confidence level threshold, the model will be considered as having a confidence level. If so, it is considered that a new fault mode has been identified.

5. The method for continuous diagnosis of rotating machinery based on future state prediction and a world-wide model according to claim 1, characterized in that, The knowledge distillation losses are as follows: ； In the formula, This represents the loss from knowledge distillation. This indicates old fault data; Represents the old world grand model. This represents the updated world model; represents the Softmax function, and KL represents the Kullback-Leibler divergence.

6. A continuous diagnostic system for rotating machinery based on future state prediction and a world-wide model, characterized in that, The system employs the continuous diagnostic method for rotating machinery based on future state prediction and a world-large model as described in any one of claims 1 to 5, comprising: The data acquisition and preprocessing module is used to acquire and preprocess multi-source data from rotating machinery to generate multi-source data sequences. The digital twin and future prediction module is used to drive the digital twin model of rotating machinery using multi-source data sequences, so as to achieve state synchronization between the digital twin model and the physical entity, and generate the first future state prediction sequence of the rotating machinery for a period of time based on the digital twin model. The world-class model diagnostic reasoning module is used to encode the multi-source data sequence and the first future state prediction sequence into a prompt vector, which is then input into the pre-trained world-class model to output diagnostic results and decision suggestions, and output the second future state prediction sequence of the rotating machinery within the same future time period. The continuous learning and model update module is used to initiate a continuous learning mechanism when a new fault mode is identified. This mechanism introduces a low-rank matrix into the world-wide model through low-rank fine-tuning and uses a total loss function, including task loss, future information consistency loss, and knowledge distillation loss, to continuously learn and update the parameters of the world-wide model. Among them, task loss and knowledge distillation loss are used to ensure that the world-wide model correctly classifies new fault data and old fault data, respectively, and future information consistency loss is used to ensure that the updated world-wide model's future state predictions for new fault data are consistent with the physical predictions of the digital twin model by constraining the second future state prediction sequence and the first future state prediction sequence. The total loss function is as follows: ； In the formula, Indicates the total loss; This represents task loss, used to ensure that the world-class model correctly classifies new faulty data. This represents the knowledge distillation loss, used to ensure that the world-large model correctly classifies old faulty data. This represents the loss of future information consistency, used to constrain the updated world-large model's prediction of the future state of new fault data to be consistent with the physical prediction of the digital twin model; and Indicates the weighting factor; The loss of future information consistency is as follows: ； In the formula, This indicates a loss of future information consistency. This indicates new fault data. Representing a digital twin model, This represents the physical prediction of the digital twin model, i.e., the first future state prediction sequence; Indicates encoding, Represents the prompt vector; Represents the parameters of the world-large model; This represents the incremental update of the parameters of the world-large model, determined by the introduced low-rank matrix; This represents the updated world model. This represents the output of the updated world model, including the updated world model's prediction of the future state of the new fault data, i.e., the second future state prediction sequence.

7. A computer device, characterized in that, include: A processor and a memory, wherein the memory stores a program or instructions that can run on the processor, and when the program or instructions are executed by the processor, implement the steps of the method for continuous diagnosis of rotating machinery based on future state prediction and world-scale model as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores programs or instructions that, when executed by a processor, implement the steps of the continuous diagnostic method for rotating machinery based on future state prediction and world-scale model as described in any one of claims 1 to 5.

Images