Diesel generating set fault detection system

Abstract

The invention discloses a diesel generating set fault detection system, belongs to the technical field of diesel generating sets, and solves the problems that a composite fault of fuel oxidation and sensor signal drifting exists in a diesel generating set, the fault has concealment and complexity, and the fault is difficult to detect. Therefore, the composite fault is difficult to check, and the diesel generating set is damaged after working. Comprising a fault analysis module, and a combustion detection module, a gas detection module and an information extraction module which are electrically connected with the fault analysis module. Flow, pressure, oxygen concentration, temperature, cylinder pressure, nitrogen oxide content, ECU correction value and fuel injection pulse width instructions are extracted, chaotic dynamics, quantum field theory and adversarial learning technologies are combined, a multi-scale feature system is constructed, causal entanglement analysis and a quantum decision-making mechanism are introduced, fuel oxidation and sensor drift are accurately distinguished, and the accuracy of fuel oxidation and sensor drift is improved. The problems that a traditional method is high in misjudgment rate and difficult in composite fault separation are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of diesel generator sets, and particularly to a fault detection system for diesel generator sets. Background Art

[0002] A diesel generator set is a power generation device that uses diesel as fuel and converts mechanical energy into electrical energy by driving a generator with a diesel engine. It mainly consists of components such as a diesel engine, a generator, a control box, a fuel tank, a starting and control storage battery, and a protection device. The diesel engine, as the power core, directly affects the efficiency and reliability of the entire generator set. The generator is responsible for converting mechanical energy into electrical energy and is the core part of power generation. The control box is used to monitor and control the entire power generation process to ensure the stable and safe operation of the generator set.

[0003] During the operation of a diesel generator set, a composite fault of fuel oxidation and sensor signal drift may occur. Fuel oxidation is caused by the oxidation reaction of unsaturated hydrocarbons in diesel under the influence of factors such as temperature, humidity, and air during storage or use, generating precipitates such as gums. These precipitates will block the components of the fuel system, affect the normal flow and injection of fuel, and result in incomplete combustion. Incomplete combustion will cause a series of problems, such as black smoke emission, increased carbon deposition, and reduced fuel efficiency.

[0004] At the same time, sensor signal drift is due to the gradual decline in the sensitivity and accuracy of key sensors such as oxygen sensors under the long-term action of high temperature and exhaust gas chemical substances. The oxygen sensor signal drift will cause the electronic control unit (ECU) of the diesel generator set to receive incorrect oxygen content information, thereby adjusting the fuel injection volume. This incorrect adjustment will further exacerbate the impact of fuel oxidation, resulting in a too rich or too lean air-fuel mixture, reduced combustion efficiency, and even possible damage to the engine.

[0005] The composite fault of fuel oxidation and sensor signal drift is characterized by concealment and complexity. Fuel oxidation is a gradual process, and its initial impact on engine performance is not obvious; sensor signal drift will also gradually accumulate over time and is difficult to detect initially. The interaction between the two makes the fault manifestation complex and increases the difficulty of detection and diagnosis. This composite fault may cause problems such as black smoke emission, insufficient power, and increased fuel consumption during the operation of the diesel generator set. In severe cases, it may damage the engine and emission system. Timely detection and handling of this composite fault are crucial for ensuring the normal operation of the diesel generator set.

[0006] Therefore, it is necessary to identify this composite fault of fuel oxidation and sensor signal drift to avoid problems such as the direct startup of the diesel generator set and damage to the diesel generator set under a series of influences.

[0007] Therefore, a fault detection system for diesel generator sets is proposed to solve or alleviate the above problems. Summary of the Invention

[0008] The purpose of the present invention is to solve the deficiencies existing in the prior art, and a fault detection system for diesel generator sets is proposed, which solves the compound faults of fuel oxidation and sensor signal drift in diesel generator sets. This compound fault has concealment and complexity, which makes it difficult to detect the compound fault, and ultimately causes the diesel generator set to be damaged after operation.

[0009] In order to achieve the above purpose, the present invention adopts the following technical solutions: A fault detection system for diesel generator sets includes a fault analysis module, and a combustion detection module, a gas detection module, and an information extraction module that are electrically connected to it. The fuel detection module, combustion detection module, gas detection module, and information extraction module collect parameters and transmit them to the fault analysis module. The fault analysis module detects compound faults of fuel oxidation and sensor signal drift through multi-dimensional feature decoupling and dynamic coupling effect modeling.

[0010] Preferably, the fault analysis module detects compound faults of fuel oxidation and sensor signal drift through multi-dimensional feature decoupling and dynamic coupling effect modeling, including the following steps: Multi-modal data synchronization and robust preprocessing, align the timestamps and suppress the noise of the original data of each module; Multi-scale dynamic feature extraction, extract the time-frequency domain features of fuel oxidation and mechanical wear through chaotic attractor reconstruction, maximum chaos index calculation, and multifractal spectrum analysis; Feature fusion inspired by quantum field theory, map the original data to a gauge field and detect topological defects in the system; Supermanifold learning and causal entanglement analysis, construct a causal path in the probability distribution space and quantify the coupling anomaly of subsystems; Adversarial continuous learning and dynamic optimization, optimize the robustness of the model through perturbation sample generation and adaptive adjustment of feature weights; Multi-criterion decision fusion, combine evidence synthesis and quantum dynamics kernel classifier to output the compound fault diagnosis result.

[0011] Preferably, the multi-modal data synchronization and robust preprocessing, align the timestamps and suppress the noise of the original data of each module, including the following steps: Use the dynamic time warping algorithm to align multi-source original data, calculate the distance between the current data point and the reference signal data point, and accumulate the minimum path cost to align the time series of each sensor signal; Build a joint denoising model of Kalman filtering and adaptive wavelet thresholding. Use the Kalman filter to perform state estimation on the signal, and perform threshold processing on the scale coefficients through wavelet decomposition to suppress high-frequency noise and retain effective signal components, where the threshold parameter is dynamically adjusted according to the noise variance.

[0012] Preferably, for the multi-scale dynamic feature extraction, through chaotic attractor reconstruction, maximum chaos index calculation, and multifractal spectrum analysis, extract the time-frequency domain features of fuel oxidation and mechanical wear, including the following steps: Determine the optimal reconstruction parameters of the chaotic attractor by calculating the average distance of neighboring points at different embedding dimensions and delay times in the time series. Based on the reconstructed attractor trajectory, evaluate the chaotic characteristics of the time series, calculate the maximum chaos index by statistically analyzing the divergence rate of adjacent trajectory points, and determine fuel oxidation when the index exceeds a preset threshold. By analyzing the probability distribution of the time series at different scales, calculate the width of the multifractal spectrum. If the spectrum width exceeds a set value, confirm that the system has a compound fault.

[0013] Preferably, for the quantum field theory-inspired feature fusion, map the original data to a gauge field and detect topological defects in the system, including the following steps: Map the sensor time series data to the potential function component in the gauge field, and detect local distortions in the system topology by calculating the curvature tensor of the gauge field. By integrating the four-dimensional space components of the gauge field curvature tensor, calculate the instanton number density. If the instanton number is non-zero, determine that the system has symmetry breaking due to a compound fault.

[0014] Preferably, for the supermanifold learning and causal entanglement analysis, construct a causal path in the probability distribution space and quantify abnormal subsystem coupling, including the following steps: In the probability distribution space, calculate the optimal transport path between the normal state and the observed state, and construct a geodesic reflecting the causal relationship. Regard each subsystem as a qubit, calculate the entanglement entropy between subsystems through the density matrix. If the entanglement entropy exceeds the statistical range of the historical reference value, determine that there is abnormal coupling between subsystems.

[0015] Preferably, for the adversarial continuous learning and dynamic optimization, optimize the model robustness through perturbation sample generation and adaptive adjustment of feature weights, including the following steps: Generate adversarial perturbation samples by maximizing the classifier loss function to enhance the model's robustness to sensor drift. Based on the second-order derivative information of the feature channels with respect to the loss function, dynamically reduce the weights of the noise-sensitive channels to suppress the risk of overfitting.

[0016] Preferably, the multi-criterion decision fusion combines evidence synthesis and the output of a quantum dynamics kernel classifier to obtain a composite fault diagnosis result, which includes the following steps: Integrate multi-source evidence of chaotic characteristics, topological defects, and coupling anomalies through the evidence synthesis rule, and calculate the joint confidence of the composite fault; Construct a classification hyperplane using a kernel function based on quantum state similarity. If the projection of a sample in the reproducing kernel Hilbert space is located in the fault region, an alarm for the composite fault is output.

[0017] The present invention has the following beneficial effects: The present invention extracts fuel flow and pressure through a fuel detection module, extracts exhaust oxygen concentration and exhaust temperature through a combustion detection module, extracts in-cylinder pressure value and nitrogen oxide content through a gas detection module, and extracts the ECU air-fuel ratio correction value and injection pulse width command of the control system in a diesel generator set through an information extraction module. Based on the above raw data, through the deep cross of chaotic dynamics, quantum field theory, and adversarial learning technologies, the present invention realizes the accurate distinction between fuel oxidation and sensor drift, constructs a spatio-temporal-frequency-domain-topological multi-scale feature system within a conventional sensor framework, and introduces causal entanglement analysis and a quantum decision mechanism, solving the technical problems of high misjudgment rate of traditional threshold methods and difficulty in separating composite faults. Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a structural block diagram of the present invention.

[0020] 1. Fuel detection module; 2. Combustion detection module; 3. Gas detection module; 4. Information extraction module; 5. Fault analysis module. Detailed Embodiments

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0022] Accordingly, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0023] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0024] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. are based on the orientation or positional relationships shown in the drawings, or the orientation or positional relationships in which the inventive product is customarily placed during use, or the orientation or positional relationships commonly understood by those skilled in the art. These are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention.

[0025] In addition, the terms "first", "second", "third", etc. are only used for descriptive distinction and should not be construed as indicating or implying relative importance.

[0026] In the description of the present invention, it should also be noted that unless otherwise clearly specified and defined, the terms "set", "installed", "connected", "coupled" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0027] A diesel generator set fault detection system, as Figure 1 shown, includes a fault analysis module 5, and a fuel detection module 1, a combustion detection module 2, a gas detection module 3, and an information extraction module 4 that are electrically connected thereto. The fuel detection module 1, the combustion detection module 2, the gas detection module 3, and the information extraction module 4 collect parameters and transmit them to the fault analysis module 5. The fault analysis module 5 detects the compound faults of fuel oxidation and sensor signal drift through multi-dimensional feature decoupling and dynamic coupling effect modeling.

[0028] The fuel detection module 1 includes a fuel flow meter and a fuel pressure sensor, the combustion detection module 2 includes an oxygen sensor and an exhaust gas temperature sensor, the gas detection module 3 includes a cylinder pressure sensor and a NOx sensor, the information extraction module 4 extracts the ECU air-fuel ratio correction value and the fuel injection pulse width command, and the fault analysis module 5 includes a processor.

[0029] In the present invention, the fuel flow and pressure are extracted by the fuel detection module 1, the exhaust gas oxygen concentration and the exhaust gas temperature are extracted by the combustion detection module 2, the cylinder internal pressure value and the nitrogen oxide content are extracted by the gas detection module 3, and the ECU air-fuel ratio correction value and the fuel injection pulse width command of the control system in the diesel generator set are extracted by the information extraction module 4. Based on the above original data, the fault analysis module 5 realizes the accurate distinction between fuel oxidation and sensor drift through the deep cross of chaotic dynamics, quantum field theory and adversarial learning technology, constructs a spatio-temporal-frequency domain-topological multi-scale feature system within the conventional sensor framework, and introduces causal entanglement analysis and quantum decision-making mechanism, solving the problems of high misjudgment rate of the traditional threshold method and difficult separation of compound faults.

[0030] Preferably, the fault analysis module 5 detects the compound faults of fuel oxidation and sensor signal drift through multi-dimensional feature decoupling and dynamic coupling effect modeling, including the following steps: Multi-modal data synchronization and robust preprocessing, performing timestamp alignment and noise suppression on the original data of the fuel pressure sensor, fuel flow meter, oxygen sensor, exhaust gas temperature sensor, cylinder pressure sensor, NOx sensor, ECU air-fuel ratio correction value and fuel injection pulse width command; Multi-scale dynamic feature extraction, extracting the time-frequency domain features of fuel oxidation and mechanical wear through chaotic attractor reconstruction, maximum chaos index calculation and multifractal spectrum analysis; Quantum field theory-inspired feature fusion, mapping the original data into a gauge field and detecting system topological defects; Supermanifold learning and causal entanglement analysis, constructing causal paths in the probability distribution space and quantifying subsystem coupling anomalies; Adversarial continuous learning and dynamic optimization, optimizing the model robustness through perturbation sample generation and feature weight adaptive adjustment; Multi-criterion decision fusion, combining evidence synthesis and quantum dynamics kernel classifier to output the compound fault diagnosis result.

[0031] In the above method steps, in the data preprocessing stage, the dynamic time warping algorithm is used to solve the multi-sensor clock deviation problem, and combined with Kalman filtering and wavelet threshold denoising, noise interference is eliminated and effective signal features are retained.

[0032] Secondly, in the dynamic feature extraction stage, by reconstructing the chaotic attractor, calculating the Lyapunov exponent, and multifractal spectrum, the nonlinear dynamic behavior caused by fuel oxidation is revealed from the time-frequency domain.

[0033] In the quantum field theory mapping stage, the sensor data is encoded as an SU(2) gauge field, and the topological defects of the system are detected through the instanton number density, breaking through the perception dimension limitation of traditional statistical methods. In the causal entanglement analysis stage, the causal geodesic is constructed in the Wasserstein space, and the abnormal coupling between subsystems is quantified by combining the quantum entanglement entropy, accurately distinguishing the correlation effect between real faults and sensor drifts.

[0034] In the adversarial learning optimization stage, perturbation samples are generated through manifold adversarial training, dynamically adjusting the feature weights to improve the robustness of the model to sensor signal drifts. Finally, in the decision fusion stage, the evidence theory is used to integrate multi-source features, and a high-confidence fault determination is achieved through a quantum dynamics kernel classifier, so as to determine with high confidence whether a composite fault of fuel oxidation + sensor signal drift occurs.

[0035] Preferably, for multi-modal data synchronization and robust preprocessing, the original data of the fuel pressure sensor, fuel flow meter, oxygen sensor, exhaust temperature sensor, cylinder pressure sensor, NOx sensor, ECU air-fuel ratio correction value, and injection pulse width command are time-stamped and noise-suppressed, including the following steps: The dynamic time warping algorithm is used to align the multi-source raw data. By calculating the distance between the current data point and the reference signal data point and accumulating the minimum path cost, the time series of each sensor signal is aligned. Specifically, the dynamic time warping algorithm is used to compensate for the sampling clock deviation of each sensor: , where is the main reference signal, is the signal to be aligned, and the minimum path cumulative cost is minimized to achieve sub-millisecond synchronization. is the cumulative cost representing from the i-th point of the reference signal x to the j-th point of the signal to be aligned y. is the Euclidean distance between two signal points. is to select the minimum cumulative cost path from the previous point to the current point to achieve the optimal alignment; A joint denoising model of Kalman filter and adaptive wavelet threshold is constructed. The signal is state-estimated by the Kalman filter, and the threshold processing is performed on each scale coefficient by wavelet decomposition to suppress high-frequency noise and retain the effective signal components. The threshold parameter is dynamically adjusted according to the noise variance. Among them, the Kalman filter-wavelet threshold joint denoising model is , KF is the state space equation: , the wavelet threshold function: , is the noise variance estimate at scale j, is the general threshold, is the estimated value of the denoised signal, is the estimate of the signal by the Kalman filter The estimated value of, J is the number of layers of wavelet decomposition, and k is the coefficient index representing wavelet decomposition, is the value of the wavelet basis function at the j-th layer and the k-th position, is the state vector representing the internal state of the system, A is the state transition matrix, and B is the control input matrix, is the control input vector, is the process noise, is the observation vector, and H is the observation matrix, is the observation noise, is the wavelet coefficient at the j-th layer and the k-th position, is the wavelet coefficient, and N is the signal length.

[0036] In the above method steps, sub-millisecond time alignment is achieved by recursively calculating the minimum cumulative distance between data points. The Kalman filter-wavelet threshold joint denoising model combines state estimation and multi-scale analysis. First, the signal trend is predicted by the Kalman filter, and then the noise frequency band is separated by wavelet decomposition and adaptively threshold processed. While retaining the mutation characteristics, high-frequency noise is suppressed, and the non-uniformly sampled data under rotational speed fluctuations can be accurately aligned. The joint denoising model improves the signal-to-noise ratio compared with single filtering techniques, especially has a significant effect on suppressing the low-frequency drift noise of oxygen sensors, ensures the reliability of subsequent feature extraction, and the calculation efficiency meets the real-time requirements.

[0037] Preferably, for multi-scale dynamic feature extraction, time-frequency domain features of fuel oxidation and mechanical wear are extracted through chaotic attractor reconstruction, maximum chaos index calculation, and multifractal spectrum analysis, including the following steps: By calculating the average distance between neighboring points at different embedding dimensions and delay times in the time series, the optimal reconstruction parameters of the chaotic attractor are determined, and the optimal embedding parameters are determined based on the improved C-C method: , select m such that reaches saturation for the first time, τ is the time delay, determined by the first minimum point of the mutual information function, is the average distance representing the embedding dimension m and the time delay τ, N is the total length of the signal, and m is the embedding dimension; Based on the reconstructed attractor trajectory, the chaotic characteristics of the time series are evaluated, and the maximum chaos index is calculated by statistically calculating the divergence rate of adjacent trajectory points. Specifically, the maximum Lyapunov exponent is estimated by the small data method: , fuel oxidation leads to an increase in the chaos degree of the system, It rises by 10% - 25%, and when the index exceeds the preset threshold, fuel oxidation is determined. Specifically, when it is determined that fuel oxidation occurs; Among them, is the final time point, is the initial time point, M is the number of trajectory pairs, is the distance of trajectory pair k at time t, is the distance of trajectory pair k at the initial time; By analyzing the probability distribution of time series at different scales, the width of the multifractal spectrum is calculated through the partition function. The partition function is , and the mass index The singularity spectrum is obtained through the Legendre transform , if the spectral width exceeds the set value, that is when, it is confirmed that there is a compound fault in the system; Among them, q is the multifractal parameter, s is the scale, is the probability density in the μ - th box at scale s.

[0038] According to the above - mentioned method steps, based on the improved C - C method, the parameters of the chaotic attractor are determined. By statistically analyzing the saturation characteristics of the distances between neighboring points under different embedding dimensions, the distortion of the phase - space structure caused by fuel oxidation is captured. The Lyapunov exponent quantifies the sensitivity of the system to the initial conditions and directly reflects the nonlinear dynamic instability caused by gum deposition. The chaotic characteristics are sensitive to early fuel oxidation and can give an early warning when the fuel supply pressure drops. The multifractal spectrum width detects the multi - scale heterogeneity of the time series and distinguishes the complex patterns of single faults and compound faults.

[0039] Preferably, the feature fusion inspired by quantum field theory maps the original data to a gauge field and detects the topological defects of the system, including the following steps: Map the sensor time - series data to the potential - function component in the gauge field, and detect the local distortion of the system topological structure by calculating the curvature tensor of the gauge field. Specifically, construct the SU(2) gauge field , and its curvature tensor is: , where is the Pauli matrix, g is the coupling constant, map the sensor data to the gauge potential, and the curvature tensor characterizes the abnormal topology of the system, is the component of the gauge field, is the partial derivative with respect to the μ direction, is the Lie bracket of the gauge field; By integrating the four - dimensional space components of the gauge - field curvature tensor, calculate the instanton number density through the Pontryagin - class integral If the instanton number is non-zero, i.e., when Q≠0, it is determined that the system has symmetry breaking due to a compound fault; Among them, is the Levi-Civita tensor, is the trace of the matrix.

[0040] Map the sensor data to the SU(2) gauge field, characterize the system topology through the curvature tensor. The gauge field mapping breaks through the dimensional limitation of traditional statistical methods, can detect the hidden topological features in the sensor signal, and the instanton number density calculation reveals non-trivial topological defects. Fuel oxidation and sensor drift will cause local distortion of the gauge potential. A non-zero instanton number indicates symmetry breaking of the system, corresponding to the coupling effect of compound faults. The instanton number density has high detection sensitivity to compound faults, strong anti-electromagnetic interference ability, and low false alarm rate in a strong noise environment.

[0041] Preferably, for supermanifold learning and causal entanglement analysis, construct a causal path and quantify the coupling anomaly of subsystems in the probability distribution space, including the following steps: In the probability distribution space, calculate the optimal transport path between the normal state and the observed state, and construct a geodesic reflecting the causal relationship. Specifically, construct a causal geodesic in the Wasserstein space: , where μ and v are the normal and fault state distributions respectively, is the joint distribution, is the Euclidean distance between two points; Regard each subsystem as a qubit, and calculate the entanglement entropy between subsystems through the density matrix If the entanglement entropy exceeds the statistical range of the historical benchmark value, it is determined that there is abnormal coupling in the subsystem; Among them, is the probability distribution, and are the basis vectors of the quantum state, is the density matrix of subsystem A, is the trace of the matrix.

[0042] In the above method steps, construct a causal geodesic in the Wasserstein space, quantify the minimum conversion cost between the normal and fault state distributions through the optimal transport theory. The causal geodesic can locate the fault propagation path, accurately distinguish the contribution of real faults and sensor drift, and the calculation of quantum entanglement entropy models the interaction of subsystems as qubit entanglement. An abnormal increase in the entropy value indicates the cross-system coupling of fuel oxidation and sensor drift. The entanglement entropy index has high specificity for compound faults and supports the tracing of the root cause of faults.

[0043] ​Preferably, adversarial continual learning and dynamic optimization are used to optimize the robustness of the model by generating perturbed samples and adaptively adjusting the feature weights, including the following steps: Generate adversarial perturbation samples by maximizing the classifier loss function , where L is the quantum cross-entropy loss function, f is the fault classifier, enhancing the model's robustness to sensor drift, is the optimal adversarial perturbation, is the norm of the perturbation, is the maximum allowable value of the perturbation, x is the input sample, and y is the true label; Based on the second-order derivative information of the loss function with respect to the feature channels, dynamically reduce the weights of the noise-sensitive channels to suppress the risk of overfitting. Specifically, dynamically adjust the feature weights based on the Hessian trace: , where is the second-order derivative of the loss function with respect to feature i, is the weight of the i-th feature at time t + 1, is the weight of the i-th feature at time t, is the learning rate, is the trace of the Hessian matrix.

[0044] In the above method steps, manifold adversarial training generates perturbed samples on the data manifold, forcing the model to learn sensor drift-invariant features. Adversarial training enables the model to maintain stable recognition of ±10% sensor drift, while the weight adjustment based on the Hessian trace identifies noise-sensitive feature channels and dynamically reduces their decision weights to suppress overfitting. The weight adaptive mechanism enables the composite fault detection accuracy to remain high even in the presence of data loss, significantly outperforming the fixed-weight model.

[0045] Preferably, multi-criterion decision fusion combines evidence synthesis and the output of a quantum dynamics kernel classifier to obtain the composite fault diagnosis result, including the following steps: Integrate multi-source evidence of chaotic features, topological defects, and coupling anomalies through the evidence synthesis rule. Specifically, synthesize the basic probability assignment through the Dempster-Shafer rule: , integrating multi-source evidence such as chaotic features, quantum field parameters, and entanglement entropy, and calculating the joint confidence of the composite fault; where is the basic probability assignment function, B and C are subsets of the evidence, and are the degrees of support of the i-th evidence for subsets B and C, and n is the number of evidence; Construct a classification hyperplane using a kernel function based on quantum state similarity. Specifically, construct the optimal separating hyperplane of the SVM classifier in the reproducing kernel Hilbert space: , the kernel function selects the quantum dynamics kernel: , where is the quantum state after the action of the time evolution operator. If the projection of the sample in the reproducing kernel Hilbert space is located in the fault area, a composite fault alarm is output; where is the output of the classifier, is the Lagrange multiplier, is the label of the i-th sample, is the kernel function, b is the bias term, N is the number of samples, and are the quantum states after the action of the time evolution operator, is the width parameter of the kernel function.

[0046] In the above method steps, the Dempster-Shafer rule fuses the uncertain evidence of chaos, topology and causal characteristics. The evidence synthesis reduces the misjudgment risk of single characteristics, and the calculation error of the composite fault confidence is low. The quantum dynamics kernel measures the similarity of samples in the Hilbert space through the evolution operator, solves the high-dimensional non-linear classification problem. The quantum kernel classifier has a higher classification accuracy for unbalanced data, supports online incremental learning, and adapts to the aging evolution of the unit.

[0047] The above is only the preferred embodiment of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A diesel generator set fault detection system, characterized in that, It includes a fault analysis module (5), and a fuel detection module (1), a combustion detection module (2), a gas detection module (3), and an information extraction module (4) that are electrically connected thereto. The fuel detection module (1), the combustion detection module (2), the gas detection module (3), and the information extraction module (4) collect parameters and transmit them to the fault analysis module (5). The fault analysis module (5) detects compound faults of fuel oxidation and sensor signal drift through multi-dimensional feature decoupling and dynamic coupling effect modeling.

2. The fault detection system for a diesel generator set according to claim 1, wherein The fault analysis module (5) detects compound faults of fuel oxidation and sensor signal drift through multi-dimensional feature decoupling and dynamic coupling effect modeling, including the following steps: Multi-modal data synchronization and robust preprocessing, performing timestamp alignment and noise suppression on the original data of each module; Multi-scale dynamic feature extraction, extracting time-frequency domain features of fuel oxidation and mechanical wear through chaotic attractor reconstruction, maximum chaos index calculation, and multifractal spectrum analysis; Quantum field theory-inspired feature fusion, mapping the original data into a gauge field and detecting topological defects of the system; Supermanifold learning and causal entanglement analysis, constructing a causal path in the probability distribution space and quantifying subsystem coupling anomalies; Adversarial continuous learning and dynamic optimization, optimizing the robustness of the model through perturbation sample generation and adaptive adjustment of feature weights; Multi-criterion decision fusion, combining evidence synthesis and quantum dynamics kernel classifier to output the compound fault diagnosis result.

3. The fault detection system for a diesel generator set according to claim 2, wherein The multi-modal data synchronization and robust preprocessing, performing timestamp alignment and noise suppression on the original data of each module, including the following steps: Using the dynamic time warping algorithm to align multi-source original data, calculating the distance between the current data point and the reference signal data point, and accumulating the minimum path cost to align the time series of each sensor signal; Constructing a joint denoising model of Kalman filter and adaptive wavelet threshold, performing state estimation on the signal through the Kalman filter, performing threshold processing on each scale coefficient through wavelet decomposition, suppressing high-frequency noise and retaining effective signal components, where the threshold parameter is dynamically adjusted according to the noise variance.

4. A diesel generator set fault detection system according to claim 2, characterized in that, The multi-scale dynamic feature extraction, extracting time-frequency domain features of fuel oxidation and mechanical wear through chaotic attractor reconstruction, maximum chaos index calculation, and multifractal spectrum analysis, including the following steps: Determining the optimal reconstruction parameters of the chaotic attractor by calculating the average distance between adjacent points at different embedding dimensions and delay times in the time series; Based on the reconstructed attractor trajectory, evaluating the chaotic characteristics of the time series, calculating the maximum chaos index by statistically analyzing the divergence rate of adjacent trajectory points, and determining fuel oxidation when the index exceeds a preset threshold; By analyzing the probability distribution of the time series at different scales, calculating the width of the multifractal spectrum, and if the spectrum width exceeds the set value, it is confirmed that there are compound faults in the system.

5. A diesel generator set fault detection system according to claim 2, wherein, The quantum field theory-inspired feature fusion, mapping the original data into a gauge field and detecting topological defects of the system, including the following steps: Mapping the sensor time series data into the potential function component in the gauge field, and detecting local distortions of the system topological structure by calculating the curvature tensor of the gauge field; By integrating the four-dimensional spatial components of the gauge field curvature tensor, the instanton number density is calculated. If the instanton number is non-zero, it is determined that the system has symmetry breaking due to compound faults.

6. The fault detection system for a diesel generator set according to claim 2, characterized in that, The supermanifold learning and causal entanglement analysis constructs causal paths in the probability distribution space and quantifies the abnormal coupling of subsystems, including the following steps: In the probability distribution space, calculate the optimal transport path between the normal state and the observed state, and construct a geodesic reflecting the causal relationship; Regard each subsystem as a qubit, calculate the entanglement entropy between subsystems through the density matrix. If the entanglement entropy exceeds the statistical range of the historical benchmark value, it is determined that there is abnormal coupling in the subsystem.

7. A diesel generator set fault detection system according to claim 2, characterized in that, The adversarial continuous learning and dynamic optimization optimize the model robustness by generating perturbation samples and adaptively adjusting the feature weights, including the following steps: Generate adversarial perturbation samples by maximizing the classifier loss function to enhance the model's robustness to sensor drift; Based on the second derivative information of the feature channels with respect to the loss function, dynamically reduce the weights of the noise-sensitive channels to suppress the risk of overfitting.

8. The fault detection system for a diesel generator set according to claim 2, wherein, The multi-criterion decision fusion combines evidence synthesis and the output of the quantum dynamics kernel classifier to obtain the compound fault diagnosis result, including the following steps: Integrate multi-source evidence of chaotic features, topological defects, and coupling anomalies through the evidence synthesis rule, and calculate the joint confidence of the compound fault; Construct a classification hyperplane using a kernel function based on quantum state similarity. If the projection of the sample in the reproducing kernel Hilbert space is located in the fault region, output a compound fault alarm.

Images