Engine fault detection method and system
By constructing a dynamic adaptive benchmark energy assessment and stroke asymmetric index, combined with time series analysis, the problem of high false alarm rate of engine under dynamic operating conditions was solved, and high-precision, low-false-alarm fault detection was achieved.
Patent Information
- Application Number
- CN202511278507.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-09
- Publication Date
- 2025-10-17
- Estimated Expiration
- 2045-09-09
AI Technical Summary
Existing engine fault detection methods are prone to false alarms when dynamic operating conditions change, and cannot effectively distinguish between real faults and normal operating condition fluctuations, resulting in reduced reliability and practicality of the diagnostic system.
A dynamic adaptive baseline energy assessment method is adopted. By calculating the similarity between the current working cycle and the historical baseline working cycle during engine operation, a dynamic baseline energy is constructed. Combined with stroke asymmetric index and time series cumulative analysis, high-precision fault diagnosis with low false alarm rate is achieved.
It effectively isolates normal vibration fluctuations caused by changes in rotational speed and load, improves the accuracy and reliability of fault detection, reduces the false alarm rate, and ensures the consistency and reliability of diagnostic results.
Smart Images

Figure CN120800805A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, and in particular to an engine fault detection method and system. BACKGROUND
[0002] As the core component of various power equipment, the reliability and safety of the engine in operation is of great importance. Online fault diagnosis technology can monitor the health status of the engine in real time during its operation, and give early warning of early faults, so as to avoid catastrophic accidents and reduce maintenance costs. Among many diagnosis technologies, fault analysis based on vibration signals is widely used due to its non-invasive and rich information.
[0003] In the prior art, a common engine fault detection method is a fixed threshold method based on vibration signal energy. This method collects the engine cylinder vibration signal through a vibration sensor, takes a single working cycle of the engine as an analysis unit, calculates the energy or other statistical characteristics of the vibration signal in each cycle, and compares it with a pre-set fixed fault threshold. When the calculated characteristic value exceeds the threshold for many times in succession, the system determines that the engine has failed.
[0004] However, there are obvious defects in actual application. The operating conditions of the engine are dynamically changing, especially in the scenarios of vehicle acceleration or deceleration or sudden load change of the generator set. Changes in operating conditions will cause a sharp fluctuation in the normal vibration level of the engine. For example, when the engine switches from low-speed idle condition to high-speed high-load condition, the healthy vibration energy of the engine itself will increase significantly.
[0005] At this time, the fixed threshold method is easy to misjudge the vibration enhancement caused by normal condition change as a fault, thereby generating a large number of false alarms. This high false alarm rate seriously reduces the credibility and practicality of the diagnosis system, making it impossible for the operator to effectively distinguish between real faults and normal condition fluctuations. The root cause lies in that this method uses a static evaluation standard to measure a dynamically changing system, and fails to establish an internal correlation between the vibration characteristics and the real-time operating state of the engine. SUMMARY
[0006] To solve the technical problems that the detection method based on the fixed threshold in the prior art cannot adapt to the dynamic change of the engine operating conditions and is prone to false alarms, the present application provides an engine fault detection method and system.
[0007] In a first aspect, the present application provides an engine fault detection method, comprising: taking any one working cycle in the engine operation process as a current working cycle, and determining a plurality of historical reference working cycles; the current working cycle / historical reference working cycle each comprises vibration signals and a crank angle of the engine. performing a weighted calculation on the vibration energy of the vibration signals of each historical reference working cycle based on the similarity between the working conditions of each historical reference working cycle and the current working cycle to determine the dynamic reference energy of the vibration signal of the current working cycle; determining the normalized energy deviation of the current working cycle in response to a comparison result between the energy of the vibration signal of the current working cycle and the dynamic reference energy; Decomposing the current working cycle into a power stroke and a compression stroke according to the crankshaft angle of the current working cycle, and determining the degree of angular velocity fluctuation of the power stroke and the degree of angular velocity fluctuation of the compression stroke respectively; By integrating the degree of angular velocity fluctuation of the power stroke and the degree of angular velocity fluctuation of the compression stroke, a stroke asymmetry index is determined to characterize the dynamic balance of the power generation and storage process inside the engine; The instantaneous risk index of the current working cycle is determined by combining the standardized energy deviation and the stroke asymmetry index. The instantaneous risk index is then cumulatively analyzed over time series to determine whether the engine has a fault.
[0008] This technical solution first establishes a dynamic baseline energy that adjusts in real time to operating conditions at the baseline setting level, completely eliminating the traditional fixed threshold. This eliminates the need for static absolute values in vibration energy assessment and transforms it into a relative value that precisely matches the current engine state. This dynamic adaptive mechanism effectively isolates normal vibration fluctuations caused by speed and load variations, a key step in addressing false alarms. Secondly, at the feature extraction level, the stroke asymmetry index is introduced. This index has a clear physical meaning and goes beyond analyzing vibration energy to delve deeper into the engine's internal power generation mechanism. By comparing the angular velocity smoothness of the power and compression strokes, it captures power output morphology variations caused by combustion anomalies such as misfire and knock. More importantly, this index cleverly utilizes normalized energy deviation for gain adjustment, making it insensitive to normal operating condition variations but highly sensitive to deviations from true fault morphology. This enables in-depth analysis from one-dimensional energy to two-dimensional morphology. Finally, at the decision confirmation level, a time series cumulative analysis mechanism is employed to verify the consistency of the instantaneous risk index. This mechanism effectively accumulates risk signals generated by real, persistent faults over time, while rapidly attenuating isolated risk values caused by random electromagnetic interference or transient data jumps, ensuring the accuracy and reliability of fault detection results. In summary, this technical solution, through the organic combination of dynamic adaptive benchmarks, deep physical features, and time series accumulation analysis, achieves high-precision, low-false-alarm, and highly robust diagnosis of engine faults.
[0009] Optionally, the plurality of historical benchmark working cycles are determined based on the following method: All historical working cycles before the current working cycle are obtained; the energy of the vibration signal of the first historical working cycle is calculated; and starting from the second historical working cycle, when the energy of the vibration signal of the historical working cycle is less than the product of the energy of the vibration signal of the previous historical working cycle and a preset coefficient, the historical working cycle is determined as a historical reference working cycle.
[0010] This technical solution constructs an online adaptive screening mechanism for a set of historical benchmark working cycles and uses a recursive relative comparison criterion for dynamic screening. It can effectively intercept energy surge signals caused by sudden or severe faults, preventing these data points representing significant abnormal conditions from entering the benchmark set, thereby ensuring the sensitivity and stability of the fault detection method in long-term operation.
[0011] Optionally, the similarity between the operating conditions of each historical benchmark operating cycle and the current operating cycle is determined by the following formula:
[0012] In the formula, For the The similarity between the working conditions of the historical benchmark working cycle and the current working cycle, and is the average speed and average load of the current working cycle, and Respectively The average speed and average load of the historical benchmark working cycle, and is the preset adjustment factor, is the natural exponential function, is the absolute value symbol.
[0013] This technical solution provides a specific and efficient method for quantifying operating condition similarity, ensuring that the calculation of dynamic benchmark energy is mainly determined by the historical benchmark working cycle that best matches the current real-time operating condition, thereby achieving accurate and smooth tracking of dynamic benchmark energy to operating condition changes, and is the accurate foundation for the establishment of the entire adaptive diagnostic framework.
[0014] Optionally, the angular velocity fluctuation degree of the power stroke and the angular velocity fluctuation degree of the compression stroke are determined based on the following manner: setting the crank angle interval corresponding to the power stroke and the compression stroke based on the power stroke of the engine, and extracting the instantaneous angular velocity sub-sequence corresponding to the power stroke and the instantaneous angular velocity sub-sequence corresponding to the compression stroke from the instantaneous angular velocity sequence of the current working cycle, respectively; wherein the instantaneous angular velocity sequence is calculated based on the crank angle and the crank speed parameter; taking the variance of the instantaneous angular velocity sub-sequence of the power stroke and the instantaneous angular velocity sub-sequence of the compression stroke as the angular velocity fluctuation degree of the power stroke and the angular velocity fluctuation degree of the compression stroke.
[0015] Optionally, the stroke asymmetry index is determined based on the following manner: The absolute difference value and the cumulative value of the angular velocity fluctuation degree of the compression stroke and the angular velocity fluctuation degree of the power stroke are calculated, and the ratio of the absolute difference value and the cumulative value is taken as the stroke asymmetry index of the current working cycle.
[0016] This technical solution realizes an adaptive normalization operation by dividing the absolute difference value by the sum of the angular velocity fluctuation degrees, effectively eliminating the influence of the angular velocity fluctuation degree caused by the overall speed and load changes of the engine, and ensuring that the stroke asymmetry index can stably measure the relative smoothness of the physical signal corresponding to the internal work during engine combustion under any working condition.
[0017] Optionally, one determination manner of the instantaneous risk index of the current working cycle is: A plurality of working cycles of the engine under a plurality of working conditions are obtained in advance; a feature combination composed of the standardized energy deviation and the stroke asymmetry index of each working cycle is obtained; a two-dimensional feature space with the standardized energy deviation as the horizontal coordinate axis and the stroke asymmetry index as the vertical coordinate axis is constructed, and a data point set composed of all feature combinations corresponding to the plurality of working cycles in the two-dimensional feature space is obtained; A cluster analysis is performed on the data point set to determine a healthy core cluster, and the centroid of the healthy core cluster is taken as a health center point; an instantaneous risk index is determined based on the distance between the data point corresponding to the feature combination of the current working cycle in the two-dimensional feature space and the health center point, wherein the instantaneous risk index and the distance are determined through a positive correlation relationship.
[0018] This technical solution introduces an adaptive risk assessment model based on unsupervised machine learning, and through a cluster analysis algorithm, a core cluster and its centroid representing a healthy state are autonomously learned and refined from a two-dimensional feature space composed of the standardized energy deviation and the stroke asymmetry index. Therefore, the instantaneous risk index is accurately quantified as the distance between the current state point and the healthy state, improving the accuracy, adaptability and automation level of the diagnosis.
[0019] Optionally, a time series cumulative analysis is performed on the instantaneous risk index to determine whether the engine has a fault, including: Calculate the decay memory fault confirmation value of the current working cycle; ;In the formula, is the attenuation memory fault confirmation value of the current working cycle, is the instantaneous risk index of the current working cycle, It is the attenuation memory fault confirmation value of the previous working cycle of the current working cycle. is a preset forgetting factor; when the decay memory fault confirmation values of a preset number of working cycles exceed a fault confirmation threshold, it is determined that the engine has a fault.
[0020] Optionally, another way to determine the instantaneous risk index of the current working cycle is: ;In the formula, is the instantaneous risk index of the current working cycle, The data point corresponding to the feature combination corresponding to the current working cycle in the two-dimensional feature space and health center points The Euclidean distance between is the normalized distance factor, is the modulation factor of the current duty cycle; where, , The centroids and The most recent suspected failure cluster; is the natural exponential function, for and The Euclidean distance between them; among them, all suspected fault clusters are all other clusters except the healthy core cluster in the results of cluster analysis.
[0021] This technical solution introduces a modulation factor on top of the basic risk assessment, enhancing the specificity and confidence of the risk assessment. This allows the diagnostic system to not only identify anomalies but also assign higher risk weights to abnormal conditions that exhibit typical failure modes, significantly improving the sensitivity of identifying real faults.
[0022] Optionally, the current working cycle / historical benchmark working cycle also includes the load of the engine, and the vibration signal is obtained through an acceleration sensor; the crankshaft angle is obtained through a crankshaft position sensor; and the load is obtained by reading an electronic control unit coupled to the engine.
[0023] In a second aspect, the present application also provides an engine fault detection system, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any one of the engine fault detection methods.
[0024] The present application has the following effects: The present application can identify the physical nature of the fault by constructing a dynamic reference energy adaptive to the working condition to realize accurate relative comparison, and analyzing the physical signal form during combustion work to determine the stroke asymmetry index. On this basis, a data-driven decision mechanism is introduced, that is, a healthy prototype is autonomously learned through unsupervised machine learning, and combined with time series cumulative analysis to distinguish between persistent faults and random disturbances, thereby improving the accuracy and reliability of the final fault detection result. BRIEF DESCRIPTION OF DRAWINGS
[0025] Figure 1 Fig. 1 is a flowchart of the present application; Figure 2 Fig. 2 is a comparison diagram of the speed signal of the current cycle and the speed signal of the historical cycle in the present application; Figure 3 Fig. 3 is a comparison diagram of the load signal of the current cycle and the load signal of the historical cycle in the present application; Figure 4 Fig. 4 is a comparison diagram of the working condition similarity of the historical cycle and the current cycle in the present application; Figure 5 Fig. 5 is a comparison diagram of the instantaneous angular velocity signal of the current cycle and the healthy cycle in the present application; Figure 6 Fig. 6 is a two-dimensional feature space constructed in the present application and the corresponding clustering analysis result diagram; Figure 7 Fig. 7 is a diagram of the process of dynamically obtaining the decay memory fault confirmation value according to the change trend of the instantaneous risk index in the present application. DETAILED DESCRIPTION
[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application.
[0027] Reference Figure 1The application provides a complete flow of an engine fault detection method. The method follows a logical order of a benchmark establishment, feature extraction and decision confirmation. First, the system tailors a dynamic health energy benchmark for the current working condition through adaptive screening of historical operation data; then, deep features representing faults are extracted from two dimensions of energy and work form; finally, the authenticity and continuity of faults are confirmed through a time sequence analysis model with time memory function.
[0028] The following describes each step in detail: Step S101: Collecting multi-modal data of the engine and determining historical benchmark working cycles.
[0029] This step is the input end and benchmark setting link of the entire diagnosis framework, and the purpose is to provide an accurate and reliable data basis for subsequent relative comparison. Unlike directly using all historical data, this step introduces an online adaptive screening mechanism to solve the potential problem that the data set used to calculate the benchmark may be contaminated by persistent fault signals.
[0030] The design logic flow of this step is as follows: First, take any working cycle in the engine operation process as the current working cycle, and synchronously collect multi-source sensor data of the engine. Then, the continuous data stream is divided into data frames with working cycles by using the crank angle signal. Then, through a recursive relative comparison criterion, the qualified cycles are screened from the historical data stream to form a historical benchmark working cycle set.
[0031] Detailed implementation description: In this embodiment, the vibration signal is obtained through the acceleration sensor installed on the engine cylinder , the crank angle signal is obtained through the high-resolution photoelectric encoder and other crank position sensors , and the load parameter is obtained by reading the ECU (Electronic Control Unit) data of the engine through the CAN (Controller Area Network) bus.
[0032] The continuous data stream is accurately divided into a complete working cycle with 720° crank angle by using the top dead center (TDC) pulse in the crank angle signal. Each working cycle contains the vibration signal, crank angle and load in the cycle, and the derivative is obtained to obtain the crank angle speed .
[0033] Then, a plurality of historical benchmark working cycles are determined based on the following methods: All historical working cycles before the current working cycle are obtained, for example, 1000 (empirical value) historical working cycles. The 1000 historical working cycles are an empirical selection balancing data coverage and real-time calculation efficiency. If the number is too small, the working condition coverage is insufficient, and if the number is too large, the system response is slowed down. Specifically, on the one hand, 1000 historical working cycles can cover common engine conditions such as idle low speed low load or high speed high load, and on the other hand, the data amount of 1000 historical working cycles does not occupy too much memory or prolong the calculation time, which meets the real-time requirement of online fault detection. For all historical working cycles before the current working cycle, the system performs a traversal screening. First, the vibration energy of the first historical working cycle is calculated, that is, the energy of the vibration signal of the first historical working cycle (the sum of squares of all amplitudes of the vibration signal). Starting from the second historical working cycle, the energy of the vibration signal thereof is calculated and compared with the energy of the vibration signal of the immediately preceding historical working cycle. If the energy of the vibration signal of the currently traversed historical working cycle is less than the product of the energy of the vibration signal of the immediately preceding historical working cycle and a preset coefficient The setting of the coefficient aims to relax the admission threshold and only exclude data with energy increasing by several times, which can be almost determined as fault, and allow fluctuation data within the normal range to enter the sample pool, thereby ensuring the data quality and sample diversity of the healthy sample pool. This simple screening can effectively prevent energy increase data caused by obvious faults from polluting the sample pool. Wherein, The core purpose of the coefficient is to intercept historical working cycles with obvious faults and retain historical working cycles with normal fluctuations. Engineering experience shows that the vibration energy fluctuation of an engine in a healthy state is usually not more than 3 times that of the previous working cycle, while faults (such as knocking and misfire) will cause vibration energy to increase dramatically, usually more than 3 times. Therefore, the coefficient is set to 3, which can filter working cycles with normal condition fluctuations and effectively capture working cycles corresponding to obvious faults.
[0034] Step S102: Calculate a dynamic reference energy according to the real-time working condition of the engine.
[0035] This step is a direct means to solve the core defects of the background technology. It combines the pure historical reference working cycles prepared in step S101 with the real-time working condition of the current engine to generate a dynamic and personalized dynamic reference energy, which changes dynamically and makes the diagnosis standard change from static to dynamic, which is the core of realizing working condition self-adaptation.
[0036] The design logic flow of this step is as follows: First, the average speed and average load of the current working cycle are obtained. Next, each historical benchmark working cycle in the set of historical benchmark working cycles is traversed and its operating condition similarity with the current working cycle is calculated. Then, using the operating condition similarity as a weight, the vibration energy of all historical benchmark working cycles is weighted averaged to obtain the dynamic baseline energy of the current working cycle.
[0037] Detailed implementation instructions: For the current cycle to be detected, the average speed is , is a whole, with an average load of , traverse all historical benchmark working cycles and calculate the similarity between the working conditions of each historical benchmark working cycle and the current working cycle.
[0038] In this embodiment, the similarity between the working conditions of the current working cycle and the historical reference working cycle satisfies the following relationship:
[0039] In the formula, For the The similarity between the working conditions of the historical benchmark working cycle and the current working cycle, and is the average speed and average load of the current working cycle, and Respectively The average speed and average load of the historical working cycle, and is the preset adjustment factor, is the natural exponential function, is the absolute value symbol. Consider two extreme cases: one is that in the current working cycle, the engine suddenly stops due to a fault (such as misfire, oil line interruption), resulting in The second is that in the current working cycle, due to sensor failure (such as sensor signal loss, circuit short circuit) or data transmission error, the load is misjudged as 0. These two extreme cases will cause the denominator to be 0 in the formula, resulting in a mathematical error and calculation crash. Therefore, in order to avoid extreme cases or 0, resulting in an unreasonable situation where the denominator is 0, so it is set uniformly here. In extreme cases, or To avoid calculation errors, set these two parameters to a very small positive number. To avoid unreasonable calculations. Specifically, by setting a very small positive number when When setting ,when When the denominator is 0, set The core of this replacement strategy is to ensure the robustness of the calculation, avoid mathematical calculation errors, and ensure that the formula can be normally operated. In the field of numerical calculation, processing the denominator as 0 with a very small positive number is a classic robust design method, and the value of is chosen because it is small enough and is a positive number close to 0, which avoids mathematical calculation errors after replacement, and fully complies with the original design logic, ensuring the correctness of the calculation results. The core of the formula is to quantify the similarity between the two through an exponential decay model that considers the influence of both speed and load. Reflects the relative difference rate or deviation degree of the average speed of the historical benchmark working cycle relative to the corresponding parameter of the current working cycle. Reflects the relative difference rate or deviation degree of the average load of the historical benchmark working cycle relative to the corresponding parameter of the current working cycle, and realizes the normalization processing of the relative difference rate or deviation degree by dividing the average speed and average load of the current cycle. The benefit of normalization is to eliminate the influence of the different dimensions and numerical ranges of speed and load. For example, a 100 RPM difference in speed and a 10 unit difference in load may not be comparable in absolute value, but by calculating the relative difference rate, they can be compared in the same dimension.
[0040] In this formula, and are preset weight coefficients or adjustment factors, respectively used to adjust the importance of speed difference and load difference in the overall similarity calculation, and are determined as follows: Calculate the coefficient of variation of speed and the coefficient of variation of load of all historical benchmark working cycles. The coefficient of variation is equal to the standard deviation divided by the average value. The coefficient of variation is usually represented by , so the coefficient of variation of speed is denoted as CV1, which reflects the fluctuation degree of speed. The greater the fluctuation degree, the greater the impact on the working condition, and vice versa. Similarly, the coefficient of variation of load is denoted as CV2, which reflects the fluctuation degree of load. The greater the fluctuation degree, the greater the impact on the working condition, and vice versa. Therefore, set , which reflects the proportion of the fluctuation degree of speed in the overall fluctuation degree. The greater the proportion, the greater the impact of the fluctuation degree of speed on the working condition in the overall fluctuation degree of speed and load, so a greater weight is set, and vice versa. Set , which reflects the proportion of the fluctuation degree of load in the overall fluctuation degree. The greater the proportion, the greater the impact of the fluctuation degree of load on the working condition in the overall fluctuation degree of speed and load, so a greater weight is set, and vice versa.
[0041] In this formula, is the weighted sum of the relative difference rate of speed and the relative difference rate of load, resulting in a comprehensive operating condition difference index. This comprehensive index comprehensively reflects the overall difference degree of the two working cycles in the two key dimensions of speed and load. The larger the difference, the larger the value of this item. By adopting an exponential decay function, a negative correlation between the overall difference degree and the operating condition similarity is established. This makes the similarity weight very sensitive to the change in operating condition difference. When the historical working cycle and the current working cycle have the same operating condition, i.e. the difference in speed and load is 0, the value in the parentheses is 0, equal to 1, and the operating condition similarity reaches the maximum value of 1. With the increase of operating condition difference, i.e. the deviation of speed or load, the absolute value of the exponential part of the exponential decay function increases, while the operating condition similarity rapidly decreases, tending to 0. This rapid decay feature precisely embodies the physical meaning of closer and more relevant, effectively amplifying the weight of samples with similar operating conditions, while rapidly suppressing the influence of samples with large operating condition difference.
[0042] Overall, the overall effect of this formula is to build a robust and sensitive operating condition similarity evaluation model. It not only simply measures the absolute gap between speed and load, but also scientifically integrates the influence of these two key parameters through normalization processing and adjustable weight factors.
[0043] In summary, by using the characteristics of the exponential decay function, this multi-dimensional operating condition difference is mapped to the operating condition similarity between them. This operating condition similarity intuitively and effectively reflects that the historical benchmark working cycle with closer average speed and average load to the current state can better represent the health level under the current state, and therefore is given a higher weight. This weight distribution is crucial for subsequent health state evaluation, fault diagnosis or performance prediction tasks, because it ensures that the model will focus on the most relevant historical data, thereby improving the accuracy and reliability of the analysis.
[0044] Please refer to Figure 2 , Figure 3 and Figure 4 , which together and step by step show the basis and results of the calculation of operating condition similarity, providing intuitive evidence for the effectiveness of this embodiment. Figure 2 The comparison results of the speed signals of the current cycle and the historical cycles are shown. The red line represents the speed signal of the current cycle, and the thin lines of other colors represent five different historical cycles. From a visual point of view, it can be clearly judged that the overall trend of the speed signals of historical cycle 1 and historical cycle 2 is most similar to that of the current cycle, indicating that their average speed difference is the smallest.Figure 3 The comparison between the load signals of the historical cycles and the load signals of the current cycle is shown. The load signals of the historical cycles 1 and 2 are closest to the current cycle. Figure 4 The figure shows the final results obtained after substituting these signal differences into the working condition similarity formula. It can be seen that although the working conditions of historical cycles 1 and 2 are visually very close to the current cycle, after precise calculation, the working condition similarity score of historical cycle 1 is the highest, while the working condition similarity values of other historical cycles with large differences in working conditions are significantly reduced.
[0045] Then, the dynamic baseline energy of the current cycle is calculated by weighting:
[0046] In this formula, is the dynamic baseline energy of the current cycle, The serial number of the historical benchmark work cycle, is the total number of historical benchmark duty cycles, For the The similarity between the operating conditions of the historical benchmark work cycle and the current work cycle is a value between 0 and 1, with larger values indicating more similar operating conditions (speed and load) between the two work cycles. This quantifies the importance or relevance of the historical benchmark work cycle, which makes the historical benchmark work cycle with the most similar operating conditions to the current work cycle dominate the calculation, while historical benchmark work cycles with significantly different operating conditions are marginalized. For the The energy of the vibration signal of a historical benchmark working cycle (the sum of the squares of all amplitudes of the vibration signal) represents the energy level of the vibration signal when the engine is running under the specific working conditions of the historical benchmark working cycle.
[0047] The numerator of this formula weights the vibration signal energy of each historical benchmark work cycle based on the degree of similarity between the current work cycle and the previous one, thus integrating the energy information of all historical benchmark work cycles. Historical benchmark work cycles with high similarity have a greater contribution to the total vibration signal energy and are included in the total at a value close to their original value. Historical benchmark work cycles with low similarity have a smaller contribution to the total vibration signal energy.
[0048] The denominator of this formula adds up all the similarities of the operating conditions for normalization. By dividing by the sum of the similarities of the operating conditions (the sum of the weights), the weighted total energy can be converted to an average energy value with practical physical significance.
[0049] In summary, the essence of the formula is to use weighted average based on operating condition similarity to construct a dynamic adaptive dynamic reference energy. This dynamic reference energy can accurately follow the normal energy fluctuation of the vibration signal caused by the change of speed and load, so as to effectively distinguish the normal operating condition change and the real equipment abnormality, and reduce the false alarm caused by the traditional fixed threshold method.
[0050] In this way, when a real fault occurs, the energy of the abnormal vibration signal will be greatly different from the dynamic reference energy, so that the fault feature is clearly highlighted, and the sensitivity and reliability of fault detection are improved.
[0051] Step S103: Extract the normalized energy deviation and stroke asymmetry index.
[0052] This step further quantifies the normalized energy deviation of the current working cycle, reflecting whether the energy of the vibration signal of the current working cycle is abnormal. At the same time, combined with the internal working mechanism of the engine, the fault features are extracted from these two dimensions, providing more abundant and more directional information for subsequent accurate decision-making, which is the key to realize from fault detection to fault classification.
[0053] The design logic flow of this step is: First, the comparison result of the energy of the vibration signal of the current working cycle and the dynamic reference energy determines the normalized energy deviation of the current working cycle; then, the crank angle of the current working cycle decomposes the current working cycle into the working stroke and the compression stroke; then, the angular velocity fluctuation degree of the working stroke and the angular velocity fluctuation degree of the compression stroke are determined; finally, the stroke asymmetry index is determined by fusing the angular velocity fluctuation degrees of the working stroke and the compression stroke.
[0054] Detailed implementation description: In this embodiment, the normalized energy deviation of the current working cycle satisfies the following relationship: Calculate the actual vibration energy of the current cycle And get the normalized energy deviation:
[0055] In this formula, is the normalized energy deviation of the current cycle, is the energy of the vibration signal of the current cycle, is the dynamic reference energy of the current cycle.
[0056] This formula defines a dimensionless normalized energy deviation, whose core role is to normalize the actual vibration energy of the current working cycle against its health benchmark in the current working condition. It intuitively reflects how many times the actual energy is the dynamic benchmark energy, and reflects the degree to which the vibration signal energy of the current working cycle deviates from its dynamic benchmark energy.
[0057] In this formula, when This means that , that is, the actual energy is almost completely consistent with the dynamic benchmark energy, which represents the most ideal and healthy running state of the engine in this working condition. When This means that , indicating that the actual energy exceeds the health benchmark in the current working condition. The size of the value quantifies the degree of excess, for example, clearly indicates that the actual energy is 1.3 times the health benchmark, that is, 30% higher, which strongly suggests that the engine may have faults such as knock, pre-ignition, etc. that cause energy to be released abnormally violently. When This means that , indicating that the actual energy fails to reach the health level it should have in the current working condition. For example, indicates that the actual energy is only 60% of the health benchmark, that is, 40% lower, which strongly suggests that the engine may have faults such as misfire, lack of fire, or insufficient combustion, etc. that cause energy output to be insufficient.
[0058] At the same time, according to the crank angle of the current working cycle, the current working cycle is divided into the working stroke and the compression stroke. For a standard four-stroke engine, the piston needs to complete four strokes of intake, compression, work, and exhaust to constitute a complete working cycle, which corresponds to two revolutions of the crankshaft, i.e. 720 degrees of crank angle. In these 720 degrees, different strokes have different functions, among which the compression stroke is the main energy consumption and force accumulation phase, while the working stroke is the only energy production and force phase. The conventional method is difficult to distinguish between normal working condition changes and real combustion faults, and the fundamental reason is that it fails to effectively utilize the inherent and stable balance relationship between the two strokes. This step is to accurately extract and compare the two strokes that constitute a unity of opposites in physical meaning, to build a feature that is not sensitive to working condition changes but highly sensitive to combustion abnormalities.
[0059] The decomposition of the current working cycle into the working stroke and the compression stroke is based on the basic working principle of the Otto cycle or the Diesel cycle of the four-stroke engine. The principle determines that, in a 720-degree cycle with the intake stroke top dead center as the 0-degree reference, the crank angle in the 180-degree to 360-degree interval must correspond to the compression stroke, and the crank angle in the 360-degree to 540-degree interval must correspond to the working stroke. The angle correspondence is determined by the mechanical structure of the engine and is constant and universal.
[0060] Therefore, for the current working cycle, the TDC (top dead center) signal is used as the reference of the angle coordinate system, the crank angle in the 180-degree to 360-degree interval is defined as the compression stroke, and the crank angle in the 360-degree to 540-degree interval is defined as the working stroke, thereby realizing the stroke decomposition of the current working cycle.
[0061] According to the crank angle of the engine at each time in the current working cycle, a crank angle sequence of the current working cycle is obtained. According to the crank angle sequence of the current working cycle and the crank speed parameter, an instantaneous angular velocity sequence of the current working cycle is obtained. The instantaneous angular velocity is the rate of change of the angle of the crank angle with respect to time, that is, the instantaneous angular velocity is the ratio of a small angle of the crank to the time spent. The final result of this process is to obtain the crank angle sequence and the instantaneous angular velocity sequence of the current working cycle, and the two sequences have the same length and are one-to-one corresponding based on time.
[0062] The instantaneous angular velocity sequence of the current working cycle is traversed, and all crank angles falling within the above compression stroke interval are extracted. According to the one-to-one correspondence between the crank angle sequence and the instantaneous angular velocity sequence, the instantaneous angular velocity values corresponding to these crank angles are obtained, and a sub-sequence of the instantaneous angular velocity corresponding to the compression stroke is constructed. Similarly, all crank angles falling within the working stroke interval are extracted, and according to the one-to-one correspondence between the crank angle sequence and the instantaneous angular velocity sequence, the instantaneous angular velocity values corresponding to these crank angles are obtained, and a sub-sequence of the instantaneous angular velocity corresponding to the working stroke is constructed.
[0063] Then, the variance of the sub-sequence of the instantaneous angular velocity corresponding to the working stroke and the variance of the sub-sequence of the instantaneous angular velocity corresponding to the compression stroke are calculated, which represent the angular velocity fluctuation degree of the working stroke and the angular velocity fluctuation degree of the compression stroke, respectively.
[0064] In this way, the analysis can be targeted within a fixed, physically meaningful window. Instead of observing the overall, generalized behavior of the cycle, the smoothness of the engine in the two key actions of charging and discharging can be directly examined. This self-referential comparison naturally cancels out the effects of overall angular velocity rise or fall due to changes in speed and load, but is extremely sensitive to signs of disruption of the balance between charging and discharging due to misfire, knock, and other faults.
[0065] Finally, based on the degree of angular velocity fluctuation of the power stroke and the degree of angular velocity fluctuation of the compression stroke, a stroke asymmetry index is determined, which characterizes the deviation of the combustion process from the standard symmetrical state.
[0066] In the present embodiment, the stroke asymmetry index is calculated by the following formula:
[0067] In this formula, is the stroke asymmetry index of the current working cycle, is the degree of angular velocity fluctuation of the power stroke of the current working cycle, which quantifies the smoothness of the engine in the discharging phase. The greater the variance, the more intense the angular velocity fluctuation, the less smooth the operation, and vice versa. is the degree of angular velocity fluctuation of the compression stroke of the current working cycle, which quantifies the smoothness of the engine in the charging phase. The greater the variance, the more intense the angular velocity fluctuation, the less smooth the operation, and vice versa, is the absolute value symbol.
[0068] A healthy engine should follow a harmonious, symmetrical rhythm in its energy consumption and generation processes. The core of each working cycle of the engine lies in the energy exchange of the two key strokes. The compression stroke is the charging or charging phase of the engine, in which energy is consumed to compress the mixture in the cylinder to provide potential energy for the power stroke. The smoothness of this process (quantified by ) reflects the stability of the charging phase. The power stroke is the discharging or discharging phase of the engine, in which the mixture burns to convert chemical energy into mechanical work to drive the piston to do work. The smoothness of this process (quantified by ) reflects the stability of the discharging phase. If the engine is in an ideal healthy state, the smoothness of the two strokes should be in a dynamic symmetrical balance, i.e. and are close in size, and any abnormal combustion (such as knock or misfire) will fundamentally disrupt this symmetry.
[0069] The formula breaks this physical symmetry and translates it into a measurable stroke asymmetry index. The numerator reflects the magnitude of the asymmetry, which is the most direct measure of the imbalance. It calculates the absolute difference between the smoothness of the engine in the output phase and the input phase. In normal cases, the better the symmetry of the two strokes, the smaller the absolute difference, the closer to 0, and the smaller the stroke asymmetry index. When the engine fails, whether it is due to the sudden increase in knock or the sudden decrease in misfire , the absolute difference will increase, the symmetry of the two strokes will be worse, and the stroke asymmetry index will be larger. The denominator is a normalized scale that adapts to the working condition. When the engine is running at high speed and high load, the overall angular velocity fluctuation will naturally be larger than when it is idling. If only the absolute difference in the numerator is considered, it may be misjudged at different working conditions. By dividing by the sum of the two angular velocity fluctuations, an adaptive normalization is achieved. It considers the magnitude of the asymmetry in the context of the current total fluctuation level, completely decoupling the overall influence of the working condition (speed, load).
[0070] Referring to Figure 5 , by comparing the instantaneous angular velocity of a healthy cycle with the current cycle, the physical connotation of the stroke asymmetry index is intuitively revealed. Figure 5 In the figure, both the left half and the right half are divided into two strokes by three vertical dashed lines. The first dashed line corresponds to 180 degrees, the second dashed line corresponds to 360 degrees, and the third dashed line corresponds to 540 degrees. The first and second dashed lines form the compression stroke, and the second and third dashed lines form the power stroke. Figure 5 The right half shows the waveform pattern of the instantaneous angular velocity of a healthy cycle. It can be seen that the peak shape is regular and the amplitude is stable, reflecting that the engine is in a healthy state, with smooth and balanced energy consumption (compression stroke) and generation (power stroke). At this time, the angular velocity fluctuation of the power stroke is close in size to the angular velocity fluctuation of the compression stroke, and the calculated stroke asymmetry index tends to 0. While Figure 5 the left half shows the waveform pattern of the instantaneous angular velocity of the current cycle, in the power stroke phase near 400° crank angle, the peak value of the angular velocity appears a significant and abnormal bulge, indicating that the engine has a violent and unstable energy release in the power phase of the cycle. This dramatic change in shape will result in a large difference between the angular velocity fluctuation of the power stroke and the angular velocity fluctuation of the compression stroke, resulting in a significant increase in the stroke asymmetry index. The stroke asymmetry index can effectively capture the imbalance of the engine's internal power shape, which is an effective feature that distinguishes from traditional energy analysis and deeply analyzes the physical nature of the fault, providing a key morphological basis for diagnosis.
[0071] In summary, the stroke asymmetry index of current work cycle is a highly refined physical index for measuring the inherent symmetry of engine core work cycle. It strictly tends to zero in healthy state; any imbalance between input and output phases caused by combustion abnormality will significantly increase its value. Therefore, the size of asymmetry index can be directly, objectively and stably used as a quantitative basis for the health degree of engine combustion process.
[0072] Step S104: Construct a two-dimensional feature space and calculate the instantaneous risk index.
[0073] This part is the pre-stage of the decision-making layer of the diagnostic system, which comprehensively utilizes the indexes extracted in step S103 to construct a two-dimensional feature space, and evaluates the instantaneous risk index according to the distribution of the current work cycle in the two-dimensional feature space.
[0074] Design logic flow: First, a two-dimensional feature space is constructed with the standardized energy deviation as the horizontal coordinate axis and the stroke asymmetry index as the vertical coordinate axis; then, a healthy core cluster is determined through clustering analysis; finally, the instantaneous risk index is determined based on the distance between the data point corresponding to the feature combination of the current work cycle in the two-dimensional feature space and the healthy center point.
[0075] Specifically: A plurality of work cycles of the engine under various working conditions (including healthy working conditions and various classic types of faults) are obtained in advance, which are processed by the foregoing steps S101-S103 to obtain the standardized energy deviation and the stroke asymmetry index of each work cycle, and the feature combination of the standardized energy deviation and the stroke asymmetry index of each work cycle is obtained; A two-dimensional feature space is constructed with the standardized energy deviation as the horizontal coordinate axis and the stroke asymmetry index as the vertical coordinate axis, then each data point of the two-dimensional feature space corresponds to a feature combination , and a data point set is obtained in the two-dimensional feature space by all the feature combinations corresponding to the plurality of work cycles.
[0076] The data points of different health states will show obvious clustering characteristics. The data point set is subjected to clustering analysis, for example, the DBSCAN (density-based spatial clustering with noise) algorithm can be used, which does not need to specify the number of clusters in advance, and can identify clusters of any shape, and can also identify outliers that do not belong to any cluster as noise, which perfectly fits the scene. The data points corresponding to the healthy state form a high-density core cluster, various fault data points may form other clusters with slightly lower density, and some random interference will be naturally identified as noise.
[0077] After the DBSCAN algorithm is executed, a plurality of clusters are output, and according to prior knowledge, a cluster containing the most data points and having a center closest to the health center point is recorded as a health core cluster, and a center of the cluster is taken as the health core point CH, so that the center of the cluster is solidified as a core parameter, and all clusters except the cluster closest to the health center point are taken as suspected fault clusters. The abscissa of the health center point represents that the actual vibration energy of the engine is completely consistent with the dynamic health baseline under the current working condition, and represents an ideal health state in the energy dimension. The ordinate represents that the angular velocity fluctuation degrees of the power stroke and the compression stroke of the engine are completely equal, and represent the perfect symmetry and balance of the internal power form of the engine.
[0078] It can be seen that the coordinate point is a reference point that best represents the ideal health state of the engine, which is determined by the physical meanings of the two core feature indicators constructed by the application. The health data points in actual operation will be closely distributed around the theoretical center point in a Gaussian distribution, and any fault will cause the data points to systematically deviate from the center in the feature space.
[0079] Then, based on the distance between the data point corresponding to the feature combination of the current working cycle in the two-dimensional feature space and the health center point, an instantaneous risk index is determined, wherein the instantaneous risk index and the distance are determined by a positive correlation formula.
[0080] Referring to Figure 6 , the figure is a two-dimensional feature space constructed by the application and a visualization presentation of the analysis result. Figure 6 The abscissa is the normalized energy deviation (NED), and the ordinate is the stroke asymmetry index (SAI), which shows that the data points of different health states exhibit obvious clustering characteristics, wherein the core health cluster is closely gathered around the theoretical health point , verifying the physical expectation that the energy deviation is small and the symmetry is good under the health state. Various suspected fault clusters and noise clusters are distributed away from the health core, and there is obvious separation between them. The data point corresponding to the current cycle in the two-dimensional space deviates from the health center point, and the red dotted line connecting the two intuitively represents the Euclidean distance between them, which is the basis for calculating the instantaneous risk index R. The two-dimensional feature space realizes effective separation of different health states of the engine, and the risk assessment method based on the distance of the data point to the health center is intuitive, quantitative and reliable, providing a solid basis for subsequent fault decision-making.
[0081] In this embodiment, the calculation of the instantaneous risk index of the current working cycle is as follows when the engine is actually running: The corresponding data point of the feature combination of the normalized energy deviation and the stroke asymmetry index of the current working cycle in the two-dimensional feature space is denoted as The instantaneous risk index of the current working cycle is defined as the normalized distance from the health center point , that is:
[0082] In this formula, is the instantaneous risk index of the current working cycle, is the corresponding data point of the feature combination of the normalized energy deviation and the stroke asymmetry index of the current working cycle in the two-dimensional feature space, is the health center point, is the Euclidean distance between and , and The greater the value is, the farther the engine state corresponding to the current working cycle is from the healthy state, and the greater the instantaneous risk index of the current working cycle is, and vice versa. is a normalized distance factor, which is a constant used to scale the range of the instantaneous risk index, and can be set as the average radius of the health core cluster.
[0083] In summary, the formula converts the fault risk into a standardized, interpretable and adaptive numerical indicator in a data-driven manner, and is a key bridge connecting advanced feature extraction and intelligent fault decision-making, which improves the accuracy and adaptability of detection.
[0084] In order to enhance the sensitivity to specific faults, a modulation factor related to the nearest fault cluster can be introduced, and another determination method of the instantaneous risk index of the current working cycle is as follows:
[0085] In this formula, is the instantaneous risk index of the current working cycle, is the Euclidean distance between and , and is a normalized distance factor, which is a constant used to scale the range of the instantaneous risk index, and can be set as the average radius of the health core cluster. is a fault modulation factor of the current working cycle, and the calculation method is as follows: , is the center of all suspected fault clusters, and The most recent suspected failure cluster; is the natural exponential function, for and The Euclidean distance between . The meaning is that when When it is not only far from the health center but also very close to the center of a certain failure mode, it is more certain that it is a real failure and is therefore given a higher risk weight. The form ensures that The closer to the center of a certain failure mode, The closer the value is to 1, the faster it decays to 0.
[0086] This cluster analysis-based approach makes the calculation of the transient risk index completely data-driven. It not only measures the degree to which the operating state corresponding to the current duty cycle deviates from a healthy state, but also, through the fault modulation factor, measures the degree to which the operating state corresponding to the current duty cycle approaches a typical fault. This makes the transient risk index assessment more intelligent, comprehensive, and accurate.
[0087] S105: Perform fault detection based on time series cumulative analysis.
[0088] This step is the final decision-making layer of the entire diagnostic framework. Its core purpose is to introduce a temporal memory mechanism based on the instantaneous risk index to effectively distinguish between real, temporally coherent faults and occasional, random interference (such as electromagnetic noise), thereby ensuring the robustness of the final diagnostic results.
[0089] The core of this step's design process involves maintaining and continuously updating a variable called the decaying memory fault confirmation value. This variable functions like a low-pass filter or leaky bucket integrator for risk signals. The decaying memory fault confirmation value is updated using the instantaneous risk index. The final fault determination is then made based on whether the decaying memory fault confirmation value consistently exceeds a threshold.
[0090] Specifically, they include: Initialization: For the Working cycle, get the The 4 consecutive (experienced value) working cycles before the first working cycle will Attenuation memory fault confirmation value of working cycles It is set to 0 to achieve initialization. Four working cycles is the minimum effective window for initializing the decay memory fault confirmation value. Because the decay memory fault confirmation value needs to be iteratively updated, four working cycles are sufficient for the decay memory fault confirmation value to iterate to a stable level through the instantaneous risk index under normal working conditions, avoiding the initial value deviation from interfering with subsequent fault judgment, and at the same time, the system startup detection time will not be delayed due to an excessively large window.
[0091] Iterative update: for the nth working cycle, the decayed memory fault confirmation value of the (n-1)th working cycle and the instantaneous risk index of the nth working cycle are used to determine the decayed memory fault confirmation value of the nth working cycle Specifically, the decayed memory fault confirmation value of the nth working cycle is:
[0092] In the formula, is the decayed memory fault confirmation value of the (n-1)th working cycle. is the instantaneous risk index of the nth working cycle, which represents the new risk information brought by the current cycle as the current risk input, is the decayed memory fault confirmation value of the (n-1)th working cycle, which represents the memory characteristics of historical risks, is the preset forgetting factor, a constant close to 1, which controls the decay rate of historical memory, The closer to 1, the stronger the memory, and here we set 0.9, which is an empirical value, 0.9 represents a strong historical memory ability, the core is to accurately match the essential characteristics of engine fault signals and the core goal of fault detection. Because in engine faults, for example, slight valve leakage, progressive cylinder misfire and slight oil line blockage, etc., are usually continuous and gradual signals, the risk value of such faults needs to be gradually accumulated through historical data of multiple running cycles to highlight from the background noise of normal operation and reach the alarm threshold; if the historical memory ability is too low, i.e. The historical data weight will decay quickly, leading to the loss of weak fault signals and the difficulty of effective accumulation, ultimately causing fault omission, which violates the fundamental purpose of detection. In short, setting 0.9 retains strong memory ability, the essence is to retain the cumulative effect of continuous fault signals, avoiding the loss of fault information due to weak memory, and improving the accuracy of fault detection. In the formula, part is the memory decay term, which represents the inertia or memory of historical risks. When an occasional disturbance pulse causes the instantaneous risk index of the current working cycle to be very high, the decayed memory fault confirmation value of the current working cycle jumps accordingly. For example,
[0093] , very low, suddenly came a very high . The formula becomes , the decay memory fault validation value jumps from 0.1 to 5.09. But in the next work cycle, due to the current work cycle instantaneous risk index regression to normal (close to 0), under the action of the forgetting factor, the current work cycle decay memory fault validation value will decay. In the next cycle , the interference disappears, , , the next cycle , . The decay memory fault validation value from decay to , and then decay to .
[0094] And when a real fault occurs, the instantaneous risk index of the current work cycle will continue to be positive in multiple consecutive work cycles. For example, real faults (such as misfire, knock) are physical and will continue to occur, so the instantaneous risk index will continue to output positive values. This allows the current work cycle decay memory fault validation value to accumulate by adding the instantaneous risk index of the current work cycle, and its value will steadily increase until it breaks through and stabilizes at a higher level. For example, is 2, , , , so it accumulates and steadily increases. The driving force for this growth comes from the current risk input.
[0095] Referring to Figure 7 , the dynamic change process of the decay memory fault validation value (DMFV) is shown, which verifies its ability to suppress interference and accumulate fault signals. Figure 7 The upper half shows the instantaneous risk index as input. At the 5th work cycle, an isolated, high-amplitude interference pulse appears. Corresponding to the DMFV value (output) in the lower half of Figure 7 , it can be seen that the DMFV jumps to 5.55, but due to the subsequent R regression to normal (close to 0), the forgetting factor Under the action of DMFV, the value of DMFV quickly decays to 5.10 in the next cycle, and fails to continue to grow. This intuitively demonstrates the ability of the mechanism to suppress sporadic interference. However, when R continues to be positive (representing a real fault) from the 11th working cycle, the DMFV value in the following figure begins to steadily accumulate, from 5.42 to 6.88, to 8.19, clearly embodying the cumulative effect of persistent fault signals. It is verified that the DMFV mechanism can effectively distinguish between isolated interference and persistent faults. By memory decay, transient noise is successfully filtered out, and by risk accumulation, real faults are accurately locked, improving the robustness and reliability of the final diagnosis result.
[0096] Let the serial number of the current working cycle in step S101 be , the serial number of the last working cycle of the current working cycle be , and the decay memory fault confirmation value of the current working cycle be:
[0097] In the formula, is the decay memory fault confirmation value of the current working cycle, is the instantaneous risk index of the current working cycle, is the decay memory fault confirmation value of the last working cycle of the current working cycle, is a preset forgetting factor, and the specific logic and the decay memory fault confirmation value of the 1st working cycle are the same.
[0098] Final fault determination: only when the decay memory fault confirmation value exceeds a preset fault confirmation threshold in a continuous preset number of working cycles, is the engine determined to have a fault. This continuous threshold exceeding criterion ensures that only stable recurring, time series confirmed abnormalities are finally recognized as real faults.
[0099] Specifically, set the fault confirmation threshold to 0.8 (empirical value), and obtain the decay memory fault confirmation values of the current working cycle and the four working cycles before the current working cycle (a total of five consecutive working cycles). If the decay memory fault confirmation values of the five consecutive working cycles all exceed 0.8, the engine is determined to have a fault, and a fault alarm is issued. If there is at least one decay memory fault confirmation value that does not exceed 0.8 among the decay memory fault confirmation values of the five consecutive working cycles, the engine is determined not to have a fault in the current working cycle.
[0100] Wherein, 0.8 is an empirical value determined by sample calibration, and the attenuation memory failure confirmation values of a plurality of working cycles in a healthy state and the attenuation memory failure confirmation values of a plurality of working cycles in a failure state are statistically large, and it is found through statistics that the attenuation memory failure confirmation values of a plurality of working cycles in a failure state will exceed 0.8, so 0.8 is set as a threshold value, and considering that engine failure (such as continuous misfire, chronic knock) has a persistent feature, and usually more working cycles are continuous, while transient interference (such as electromagnetic noise) usually has fewer working cycles, therefore, 5 continuous working cycles are set as a judgment condition, which can exclude transient interference and confirm the authenticity of failure, when the attenuation memory failure confirmation values of 5 continuous working cycles all exceed 0.8, it indicates that the anomaly is a persistent physical failure, rather than random interference, which meets the diagnostic principle of engineering failure that needs to be confirmed to be repeated, and this mechanism ensures that only stable and repeated anomalies are confirmed as failure to issue a failure alarm.
[0101] The engine fault detection system of the present application comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the operations of steps S101-S105 to accurately detect the failure of the engine.
[0102] In summary, the present application constructs a complete, scientific and reliable engine fault diagnosis scheme through the organic combination of the above steps.
[0103] The above only describes the preferred embodiments of the present application and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A method for detecting engine failure, characterized in that: include: Taking any working cycle during the operation of the engine as the current working cycle, and determining a plurality of historical benchmark working cycles; The current working cycle / historical reference working cycle both include the vibration signal and crankshaft angle of the engine; Performing a weighted calculation on the vibration energy of the vibration signal of each historical reference working cycle according to the similarity between the working conditions of each historical reference working cycle and the current working cycle to determine the dynamic reference energy of the vibration signal of the current working cycle; determining a normalized energy deviation of the current working cycle in response to a comparison result of the energy of the vibration signal of the current working cycle and the dynamic reference energy; Decomposing the current working cycle into a power stroke and a compression stroke based on the crankshaft angle of the current working cycle, and determining the degree of angular velocity fluctuation of the power stroke and the degree of angular velocity fluctuation of the compression stroke respectively; fusing the degree of angular velocity fluctuation of the power stroke and the degree of angular velocity fluctuation of the compression stroke to determine a stroke asymmetry index for characterizing the dynamic balance of the power generation and storage processes within the engine; The instantaneous risk index of the current working cycle is determined by combining the standardized energy deviation and the stroke asymmetry index. The instantaneous risk index is then cumulatively analyzed over time series to determine whether the engine has a fault.
2. The engine failure detection method according to claim 1, characterized in that: The multiple historical baseline duty cycles are determined based on the following: Get all historical work cycles before the current work cycle; Calculate the energy of the vibration signal of the first historical working cycle; Starting from the second historical working cycle, when the energy of the vibration signal of the historical working cycle is less than the product of the energy of the vibration signal of the previous historical working cycle and a preset coefficient, the historical working cycle is determined as a historical reference working cycle.
3. The engine failure detection method according to claim 1, characterized in that: The similarity between the operating conditions of each historical benchmark working cycle and the current working cycle is determined by the following formula: ; In the formula, For the The similarity between the working conditions of the historical benchmark working cycle and the current working cycle, and is the average speed and average load of the current working cycle, and Respectively The average speed and average load of the historical benchmark working cycle, and is the preset adjustment factor, is the natural exponential function, is the absolute value symbol.
4. The engine failure detection method according to claim 1, characterized in that: The degree of angular velocity fluctuation during the power stroke and the degree of angular velocity fluctuation during the compression stroke are determined based on the following method: The crankshaft angle intervals corresponding to the power stroke and the compression stroke are set based on the power process of the engine, and the instantaneous angular velocity subsequence corresponding to the power stroke and the instantaneous angular velocity subsequence corresponding to the compression stroke are extracted from the instantaneous angular velocity sequence of the current working cycle. The instantaneous angular velocity sequence is calculated based on the crankshaft angle and crankshaft speed parameters. The variances of the instantaneous angular velocity subsequence of the power stroke and the instantaneous angular velocity subsequence of the compression stroke are respectively used as the angular velocity fluctuation degrees of the power stroke and the angular velocity fluctuation degrees of the compression stroke.
5. The engine failure detection method according to claim 4, characterized in that: The stroke asymmetry index is determined based on the following: The absolute difference and cumulative value of the angular velocity fluctuation degree of the compression stroke and the angular velocity fluctuation degree of the power stroke are calculated, and the ratio of the absolute difference to the cumulative value is used as the stroke asymmetry index of the current working cycle.
6. The engine failure detection method according to claim 1, characterized in that: One way to determine the instantaneous risk index of the current work cycle is: Acquire multiple working cycles of the engine under various operating conditions in advance; obtain a feature combination consisting of a standardized energy deviation and a stroke asymmetry index for each working cycle; construct a two-dimensional feature space with the standardized energy deviation as the horizontal axis and the stroke asymmetry index as the vertical axis, and obtain a data point set consisting of all feature combinations corresponding to multiple working cycles in the two-dimensional feature space; perform cluster analysis on the data point set to determine a healthy core cluster, and use the centroid of the healthy core cluster as the healthy center point; An instantaneous risk index is determined based on the distance between the data point corresponding to the feature combination corresponding to the current working cycle in the two-dimensional feature space and the health center point, wherein the instantaneous risk index and the distance are determined by a positive correlation relationship.
7. The engine failure detection method according to claim 1, characterized in that: Perform time series cumulative analysis of the instantaneous risk index to determine whether the engine has a fault, including: Calculate the decay memory fault confirmation value of the current working cycle; ; In the formula, is the attenuation memory fault confirmation value of the current working cycle, is the instantaneous risk index of the current working cycle, It is the attenuation memory fault confirmation value of the previous working cycle of the current working cycle. is the preset forgetting factor; When the decay memory fault confirmation values of a preset number of consecutive working cycles all exceed a fault confirmation threshold, it is determined that a fault exists in the engine.
8. The engine failure detection method according to claim 6, characterized in that: Another way to determine the instantaneous risk index of the current working cycle is: ;In the formula, is the instantaneous risk index of the current working cycle, The data point corresponding to the feature combination corresponding to the current working cycle in the two-dimensional feature space and health center points The Euclidean distance between is the normalized distance factor, is the modulation factor of the current duty cycle; where, , The centroids of all suspected fault clusters and The most recent suspected failure cluster; is the natural exponential function, for and The Euclidean distance between them; among them, all suspected fault clusters are all other clusters except the healthy core cluster in the results of cluster analysis.
9. The engine failure detection method according to claim 1, characterized in that: The current working cycle / historical reference working cycle also includes the load of the engine, and the vibration signal is obtained by the acceleration sensor; the crankshaft angle is obtained by the crankshaft position sensor; The load is obtained by reading the electronic control unit coupled to the engine.
10. An engine fault detection system, characterized in that: The engine fault detection system includes a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the steps of the engine fault detection method according to any one of claims 1 to 9.
Citation Information
Patent Citations
Engine misfire failure diagnosis method based on cylinder pressure estimating and manifold learning
CN106844922A
Engine misfire coordinated diagnosis method for hybrid electric vehicle
CN109263656A
Diesel engine fault monitoring and early warning method and device
CN111157249A
Engine misfire fault detection method and device, medium and equipment
CN117248997A
Real-time fault monitoring method for four-stroke diesel engine
CN117288476A
