A diesel engine fault early warning method based on multi-source data fusion

Abstract

The present application relates to the technical field of electric digital data processing, in particular to a diesel engine fault early warning method based on multi-source data fusion, comprising: obtaining a historical fault sample set and target window data of a diesel engine; obtaining fault association between historical running state data of each dimension and each type of fault; obtaining a single-dimension BPA function; fusing the single-dimension BPA function to obtain a fused BPA function; and obtaining a fault prediction result of the diesel engine. The present application introduces DS evidence theory to fuse the multi-dimension data collected during the operation of the diesel engine, fully utilizes the response capability of various sensors to fault characteristics under different working conditions, evaluates the representation strength of each dimension and each type of fault in the historical data, constructs a dynamic BPA function, and enhances the pertinence and sensitivity of the early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a diesel engine fault early warning method based on multi-source data fusion. Background Art

[0002] With the widespread adoption of diesel engines in construction machinery, heavy-duty trucking, ship propulsion, and distributed power generation, their operational safety and stability are attracting increasing attention. During long-term, continuous operation, diesel engines are susceptible to a variety of factors, including fuel quality fluctuations, mechanical wear, and cooling system anomalies, leading to the risk of performance degradation or sudden failure. At the same time, modern diesel engines are equipped with an increasing number of sensors and controllers, providing more comprehensive operating data and laying the foundation for operational status monitoring and analysis. In this context, leveraging this multi-source operational data to achieve refined perception and intelligent management of diesel engine operating status has become a key development direction for the industry.

[0003] Existing Problem: Existing diesel engine fault warning systems often rely on single-sensor threshold settings or rule-based matching to determine anomalies. These methods typically use a key engine indicator (such as exhaust temperature, cylinder pressure, or vibration acceleration) exceeding a set threshold as a fault criterion. These methods struggle to capture subtle changes in the early stages of a fault and often result in false, delayed, or missed alarms when operating under multiple operating conditions, with varying loads, or with minor anomalies. Summary of the Invention

[0004] The present invention provides a diesel engine fault early warning method based on multi-source data fusion to solve the existing problems.

[0005] The diesel engine fault early warning method based on multi-source data fusion of the present invention adopts the following technical solutions:

[0006] An embodiment of the present invention provides a diesel engine fault warning method based on multi-source data fusion, comprising: obtaining a historical fault sample set and target window data of a diesel engine; wherein the target window data includes operating status data at the current moment and operating status data of a preset time length selected from the current moment forward; the historical fault sample set and the target window data are both multidimensional data; based on the historical fault sample set, obtaining the fault correlation between the historical operating status data of each dimension and each type of fault; based on the data difference between the historical operating status data of each dimension and the target window data, obtaining a single-dimensional BPA function; based on the correlation between the historical operating status data of each dimension, fusing the single-dimensional BPA function to obtain a fused BPA function; based on the fused BPA function, obtaining a fault prediction result of the diesel engine.

[0007] Furthermore, the obtaining of the fault correlation between the historical operating status data of each dimension and each type of fault includes: obtaining a Pearson correlation coefficient between the historical operating status data of each dimension and each type of fault.

[0008] Furthermore, the obtaining of the fault correlation between the historical operating status data of each dimension and each type of fault also includes: for the same type of fault, obtaining a fluctuation penalty term based on the standard deviation of the historical operating status data of each dimension under the same fault severity; wherein the fault severity includes no fault occurrence, initial fault occurrence, aggravated fault and equipment failure caused by the fault; based on the fluctuation penalty term, the Pearson correlation coefficient is corrected to obtain the fault correlation between the historical operating status data of each dimension and each type of fault.

[0009] Furthermore, the obtaining of a single-dimensional BPA function based on the data difference between the historical operating status data and the target window data in each dimension includes: obtaining the degree of direct support of the target window data in each dimension for each type of fault based on the dynamic time warping (DTW) distance between the historical operating status data and the target window data in each dimension; and correcting the degree of direct support of the target window data in each dimension for each type of fault based on the fault correlation to obtain a single-dimensional BPA function.

[0010] Furthermore, after obtaining the single-dimensional BPA function, the method further includes: performing normalization processing on the single-dimensional BPA function.

[0011] Furthermore, the single-dimensional BPA functions are fused based on the correlation between the historical operating status data of each dimension to obtain a fused BPA function, including: fusing the single-dimensional BPA functions of two different dimensions to obtain an initial fused BPA function; based on the conflict degree of two different dimensions, the initial fused BPA function is corrected to obtain a fused BPA function; wherein the conflict degree is used to characterize the degree of conflict between the single-dimensional BPA functions of different dimensions when supporting different types of faults.

[0012] Furthermore, the fusing of the single-dimensional BPA functions of two different dimensions to obtain an initial fused BPA function includes: obtaining a fusion weight of each dimension; wherein the fusion weight is used to characterize the importance of the single-dimensional BPA function of the corresponding dimension when it is fused with the single-dimensional BPA function of other dimensions; based on the fusion weight, fusing the single-dimensional BPA functions of two different dimensions to obtain an initial fused BPA function.

[0013] Furthermore, the initial fusion BPA function is corrected based on the conflict degrees of two different dimensions to obtain a fusion BPA function, including: correcting the conflict degrees of two different dimensions based on the fusion weights of each dimension to obtain a corrected conflict degree; and correcting the initial fusion BPA function based on the corrected conflict degree to obtain a fusion BPA function.

[0014] Furthermore, obtaining the fusion weight of each dimension includes: obtaining the negative value of the correlation between the historical operating status data of each dimension and the historical operating status data of other dimensions, and normalizing the negative value of the correlation; for each dimension, averaging the negative value of the correlation between the current dimension and other dimensions after normalization to obtain the fusion weight of each dimension.

[0015] Furthermore, obtaining the fault prediction result of the diesel engine based on the fused BPA function includes: obtaining the support degree of the target window data for each type of fault based on the fused BPA function; and determining that the diesel engine has a corresponding type of fault when the support degree of the corresponding type of fault is greater than a preset decision threshold.

[0016] The beneficial effects of the technical solution of the present invention are:

[0017] In summary, in an embodiment of the present invention, a historical fault sample set and target window data are obtained for a diesel engine. Based on the historical fault sample set, fault correlations between historical operating status data of each dimension and various types of faults are obtained. A single-dimensional BPA function is obtained based on the data differences between the historical operating status data of each dimension and the target window data. Based on the correlations between the historical operating status data of each dimension, the single-dimensional BPA functions are fused to obtain a fused BPA function. Based on the fused BPA function, a diesel engine fault prediction result is obtained. Thus, the present invention incorporates the Dempster-Shafer evidence theory to fuse the multi-dimensional data collected during diesel engine operation, fully utilizing the responsiveness of various sensors to fault characteristics under different operating conditions. By evaluating the representation strength of each dimension in the historical data and each type of fault, a dynamic BPA function is constructed, thereby enhancing the pertinence and sensitivity of the early warning. Furthermore, by improving the fusion mechanism and introducing a correlation correction strategy between data dimensions, the impact of redundant interference on diagnostic results is effectively reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A schematic flow chart of a diesel engine fault warning method based on multi-source data fusion provided by an embodiment of the present invention;

[0020] Figure 2 A schematic diagram of the process of fusing single-dimensional BPA functions provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] To further illustrate the technical means and effectiveness of the present invention in achieving its intended objectives, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation, structure, features, and effectiveness of a diesel engine fault warning method based on multi-source data fusion proposed by the present invention. In the following description, references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0022] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0023] The following describes in detail a specific solution for a diesel engine fault early warning method based on multi-source data fusion provided by the present invention with reference to the accompanying drawings.

[0024] See also Figure 1 , which shows a diesel engine fault early warning method based on multi-source data fusion provided by one embodiment of the present invention, comprising:

[0025] Step S110: Obtain the historical fault sample set and target window data of the diesel engine; wherein the target window data includes the operating status data at the current moment and the operating status data of a preset time length selected from the current moment forward; the historical fault sample set and the target window data are both multidimensional data.

[0026] It should be noted that historical operating status data corresponding to various typical faults that have occurred in the diesel engine's history, such as injection system failure, intake system blockage, cooling system abnormality, and incomplete combustion, can be collected. This collected historical operating status data can cover key monitoring indicators of the diesel engine, such as in-cylinder pressure, injection timing, speed, exhaust temperature, intake and exhaust pressure, vibration data, oil pressure, and fuel consumption rate. This historical operating status data is multi-source data, encompassing multiple dimensions. Sources may include controller area networks (CAN), industrial sensors, vibration acquisition modules, thermal imaging devices, and fault recording systems. It exhibits temporal synchronization and physical correlation, providing a rich feature foundation for subsequent data modeling and fault identification. Furthermore, to enhance the practicality and robustness of the diesel engine fault early warning method based on multi-source data fusion, special attention should be paid to the data's changing characteristics prior to the occurrence of different faults. The collected historical operating status data can include data from normal operation, the initial stage of a fault, the process of fault aggravation, and complete failure. At the same time, combined with fault injection testing or historical fault condition backtracking, a representative fault sample set is constructed. Through data collection and construction, subsequent steps will use this data to learn fault evolution patterns and key discriminant features, providing data support for accurate and timely fault warnings.

[0027] It should be noted that the above historical fault sample set and target window data can be pre-processed by denoising and other operations. After pre-processing the multi-dimensional data, the data obtained by multiple sensors are recorded as to , and then obtain real-time data during the subsequent diesel engine operation process to provide early warning of diesel engine failure.

[0028] Step S120: Based on the historical fault sample set, the fault correlation between the historical operating status data of each dimension and each type of fault is obtained.

[0029] It should be noted that during diesel engine operation, different fault types have different sensitivities to different data dimensions. Some dimensional data changes significantly under certain faults, while other faults may show almost no response. If all dimensions are weighted equally without discrimination, weakly correlated or even irrelevant data will interfere with the overall judgment, reducing fusion efficiency and accuracy. For example, oil temperature data is very sensitive to changes in lubrication system faults (such as oil line blockage) because the oil film becomes thinner when the oil pressure decreases and the oil temperature rises. However, oil temperature has a weak response to mechanical faults such as valve sticking. In this case, even if the oil temperature is normal, the valve problem cannot be ruled out. Therefore, before performing a fault warning, the historical representation strength between each data dimension and each fault type can be evaluated to avoid incorporating meaningless data into the subsequent fusion process.

[0030] Preferably, in one embodiment of the present invention, the above step S120 may include: obtaining the Pearson correlation coefficient between the historical operating status data of each dimension and each type of fault.

[0031] It should be noted that the above-mentioned Pearson correlation coefficient is a commonly used statistical tool for measuring the degree of linear correlation between two variables. Its value range is between -1 and 1. When the Pearson correlation coefficient is 1, it represents that there is a complete positive linear correlation between the two variables, that is, when one variable increases, the other variable also increases proportionally. When the Pearson correlation coefficient is 0, it indicates that there is no linear correlation between the two variables. When the Pearson correlation coefficient is -1, it indicates that there is a complete negative linear correlation between the two variables, that is, when one variable increases, the other variable decreases proportionally. It can be understood that the Pearson correlation coefficient is a relatively mature and well-known technology. Please refer to the relevant technology for its calculation method, and the embodiments of the present invention will not be repeated.

[0032] It's important to note that different fault types have varying sensitivities to different data dimensions. By calculating the Pearson correlation coefficient, we can quantify the degree of linear correlation between each data dimension and a specific fault type. Furthermore, if certain data dimensions have little or no correlation with a particular fault type, the data from these dimensions may not be effective in fault early warning, increasing computational complexity and introducing noise. Correlation analysis can identify data dimensions that are highly correlated with fault types, thereby optimizing data utilization and improving the efficiency and accuracy of fault early warning.

[0033] Preferably, in one embodiment of the present invention, the above step S120 may further include: for the same type of fault, based on the standard deviation of the historical operating status data of each dimension under the same fault severity, obtaining a fluctuation penalty term; wherein the fault severity includes the fault not occurring, the fault initial occurrence, the fault aggravation, and the fault causing equipment failure; based on the fluctuation penalty term, correcting the Pearson correlation coefficient to obtain the fault correlation between the historical operating status data of each dimension and each type of fault. This embodiment is for example:

[0034] In multiple historical time periods, the historical operating status data is manually identified and labeled, and the historical time periods are divided into four different states: normal operation state, initial fault state, aggravated fault state and complete failure state. For dimension data, if its data changes significantly with the severity of the state, the correlation between the dimension data and the corresponding fault is strong. Dimensional data and Fault correlation between fault types for:

[0035]

[0036] in, Indicates the Pearson correlation coefficient between two sequences; Indicates the The historical operating status data of the dimension is The mean of the data in the time period; It represents the standard deviation function; Indicates the The corresponding time period Severity of the fault type. The specific values ​​are as follows:

[0037]

[0038] According to the above formula, when the correlation coefficient When it is larger, The historical operating status data of the dimension and the The correlation between different types of failures is higher. Here, the correlation can be 1 or -1. When the correlation approaches -1, it means that the decrease in the value of the corresponding dimension data will be accompanied by an increase in the probability of failure.

[0039] The denominator in the above formula is the fluctuation penalty term. The fluctuation penalty term mainly considers the data volatility corresponding to the current dimension under the same fault severity. When the data fluctuation is large under the same fault severity, the data in this dimension is not only correlated with a single diesel engine fault, so its association with a single fault type needs to be reduced.

[0040] It should be noted that when a data dimension fluctuates significantly under the same fault severity, the data in that dimension is not simply caused by a single fault but may also be affected by other factors (such as changes in operating conditions and sensor noise). The fluctuation penalty term takes data volatility into account and appropriately penalizes dimensions with large fluctuations, reducing their influence on fault warning results and thus improving warning accuracy. Furthermore, the fluctuation penalty term can more accurately reflect the correlation between data dimensions and fault types. If the data of a certain dimension fluctuates significantly under a specific fault, it indicates that this dimension may be related not only to the target fault but also to other factors. By introducing the fluctuation penalty term, the direct correlation between the dimension and the fault can be more accurately assessed, avoiding misjudgments caused by data fluctuations.

[0041] Step S130: obtaining a single-dimensional BPA function based on the data difference between the historical operating status data of each dimension and the target window data.

[0042] Preferably, in one embodiment of the present invention, step S130 may include: obtaining the degree of direct support of the target window data in each dimension for each type of fault based on the dynamic time warping (DTW) distance between the historical operating status data in each dimension and the target window data; and correcting the degree of direct support of the target window data in each dimension for each type of fault based on fault correlation to obtain a single-dimensional BPA function. After obtaining the single-dimensional BPA function, the step further includes: normalizing the single-dimensional BPA function.

[0043] It should be noted that the DTW (Dynamic Time Warping) distance is a metric used to measure the similarity between two time series. It is primarily used to align two time series on the time axis to minimize their similarity. It allows for nonlinear scaling of time series, thus better matching sequences with similar shapes but warped timelines. The DTW distance is a well-established and well-known technique. Its specific calculation method is described in related literature and will not be further detailed in this embodiment.

[0044] It should be noted that the above-mentioned BPA function (Basic Probability Assignment) is a core parameter in the DS (Dempster-Shafe) evidence theory. The BPA function is used to represent the confidence of each hypothesis (or combination of hypotheses). It assigns confidence to each subset in the identification framework rather than to a single element. For example, the BPA function can represent the degree of support of a certain sensor data for a specific fault type. DS evidence theory is a mathematical theory for processing uncertain information and multi-source data fusion. In addition to the BPA function, DS evidence theory also includes the following two core concepts: (1) Identification framework: It is a set of all possible hypotheses, each hypothesis represents a possible fault type (such as fuel system failure, cooling system abnormality, etc.). In this embodiment of the present invention, the identification framework can be a set of all possible fault types of the diesel engine. (2) Evidence fusion rule: It is the core part of the DS evidence theory and is used to fuse evidence from different information sources (such as different sensors or features). Through the Dempster rule, multiple BPA functions can be merged into a comprehensive BPA function, thereby obtaining a more reliable decision result.

[0045] In DS evidence theory, the BPA function is a form of confidence assignment given to a proposition (such as "fault type A occurs"), as follows: Indicates the The historical operating status data of the dimension is of great significance to the proposition “ In most engineering practices, traditional BPA function construction methods usually rely on single variable threshold division or fuzzy rules. For example, a threshold interval is set for a sensor. When it is greater than the threshold, , otherwise there is . In the diesel engine fault warning scenario, due to the presence of multiple working conditions disturbances, the diesel engine operating status is affected by complex factors such as cold start, high load, and ambient temperature changes, which will cause the value of the same sensor to fluctuate greatly under different conditions, thereby causing the fixed threshold to fail. In addition, for many faults such as the initial stage of bearing wear and slight blockage of the injector, the early data changes of the fault are extremely subtle, and it is difficult to judge only by the threshold. Therefore, the construction of the BPA basic probability distribution function is considered based on the difference between the target window data and the historical operating status data in the historical time period. For the first In terms of dimension, take the current moment as the center and select forward The data change sequence at each moment is used as the target window data, and then compared with the historical operation status data of multiple time periods with annotations. Dimensional data on the proposition " Type of failure occurs" direct support level for:

[0046]

[0047] in, Indicates the number of time periods marked in history; Indicates the The corresponding time period The severity of the fault type; Indicates the The data change sequence of dimension data in the current time period; Indicates the The historical running status data of the dimension is in the history Data change sequence under time period; represents the DTW distance between two sequences.

[0048] Taking into account the use of Dimensional data on the proposition " When judging the occurrence of a type of fault, the magnitude of the fault correlation is different. When the fault correlation is small, even if the difference between the current data and the historical operating status data is small, it cannot prove that the first The probability of type failure is higher. Dimensional data on the proposition " Type of failure occurs" direct support level for:

[0049]

[0050] in, Represents the normalization function.

[0051] according to The calculation method can obtain the The historical operating status data of each dimension is used to determine the support level of each type of fault, and then the BPA function of the dimension is obtained. :

[0052]

[0053] in, Indicates the number of fault types; Indicates historical operating status data with no abnormalities found.

[0054] Step S140: Based on the correlation between the historical operating status data of each dimension, the single-dimensional BPA functions are fused to obtain a fused BPA function.

[0055] See also Figure 2 Preferably, in one embodiment of the present invention, the above step S140 may include:

[0056] Step S141: Fusing two single-dimensional BPA functions of different dimensions to obtain an initial fused BPA function;

[0057] Step S142: Based on the conflict degree of two different dimensions, the initial fusion BPA function is modified to obtain a fusion BPA function; wherein the conflict degree is used to characterize the conflict degree of the single-dimensional BPA functions of different dimensions when supporting different types of faults. For example, this embodiment:

[0058] For the two different dimensions of data about the support level of each type of fault, record the With the After the dimension data corresponds to the BPA function, the Type failure support for:

[0059]

[0060]

[0061] in, Indicates the degree of conflict.

[0062] It should be noted that the above-mentioned conflict degree (Conflict) is an important concept in DS evidence theory, which is used to measure the degree of contradiction between different information sources. Specifically, the conflict degree measures the inconsistency of different information sources (such as different sensors or features) in supporting the same hypothesis (such as a certain fault type). In the above-mentioned diesel engine fault early warning method based on multi-source data fusion, the conflict degree can reflect the degree of contradiction between different data sources. If the degree of support for the same fault type from different data sources varies greatly, it means that there is a high degree of conflict between these data sources, which may be because some data sources are affected by noise or abnormal operating conditions. In addition, in traditional DS evidence theory, multiple highly correlated dimensional data may jointly strongly support a certain fault type, resulting in too high confidence in the fault type. The conflict degree can help identify this situation and avoid misjudgment due to over-fusion of redundant data.

[0063] It should be noted that: in a typical diesel engine system, multiple types of sensors are usually deployed, and these sensors cover multiple functional subsystems (fuel system, lubrication system, cooling system, emission system, etc.). In actual applications, these sensors are often not completely independent of each other. For example, a rise in oil temperature is usually accompanied by a drop in oil pressure. For example, a delay in injection leads to incomplete combustion, which increases smoke density. However, when performing multi-source information fusion, the traditional DS evidence theory assumes that each sensor information source is independent of each other and has equivalent credibility. All BPA functions are fused one by one through the Dempster rule. This processing method is effective under ideal conditions, but in actual diesel engine systems, there will be multiple highly correlated dimensions that strongly support a certain fault type. DS fusion will strengthen this support, thereby overestimating the probability of a certain fault. Therefore, the embodiment of the present invention considers introducing the correlation between dimensions on the basis of the original Dempster rule to improve the traditional DS fusion method. The improvement scheme is as follows:

[0064] Preferably, in one embodiment of the present invention, the above-mentioned step S141 fuses two single-dimensional BPA functions of different dimensions to obtain an initial fused BPA function, including: obtaining a fusion weight of each dimension; wherein the fusion weight is used to characterize the importance of the single-dimensional BPA function of the corresponding dimension when it is fused with the single-dimensional BPA function of other dimensions; based on the fusion weight, fusing the two single-dimensional BPA functions of different dimensions to obtain an initial fused BPA function.

[0065] It should be noted that the aforementioned fusion weights can be used to adjust the influence of each dimension's data in the final decision. By calculating the fusion weights, we can determine which dimensions are more representative of a specific fault warning, thereby optimizing data utilization and improving the accuracy of the warning. In addition, when there is a high degree of correlation between multiple data dimensions, the information they provide may be redundant. Fusion weights can be used to reduce the influence of these highly correlated dimensions, preventing redundant data from excessively influencing the fusion results. Fusion weights can also help identify and reduce the influence of data dimensions that are significantly affected by noise or abnormal operating conditions, thereby enhancing the robustness of the model under complex operating conditions.

[0066] Preferably, in one embodiment of the present invention, the above-mentioned step S142 corrects the initial fusion BPA function based on the conflict degrees of two different dimensions to obtain the fusion BPA function, including: correcting the conflict degrees of the two different dimensions based on the fusion weights of each dimension to obtain the corrected conflict degree; and correcting the initial fusion BPA function based on the corrected conflict degree to obtain the fusion BPA function. This embodiment is for example:

[0067] According to the obtained fusion weights of each dimension, the improved With the After the dimension data corresponds to the BPA function, the Type failure support for:

[0068]

[0069] in, Indicates the improved conflict degree:

[0070]

[0071] Then we can get the improved With the Dimensional data corresponding to BPA function ,According to the above method, the subsequent dimensional data are fused in turn to obtain the final fused BPA function.

[0072] Preferably, in one embodiment of the present invention, obtaining the fusion weight of each dimension includes: obtaining the negative value of the correlation between the historical operating status data of each dimension and the historical operating status data of other dimensions, and normalizing the negative value of the correlation; for each dimension, averaging the negative value of the normalized correlation between the current dimension and other dimensions to obtain the fusion weight of each dimension. This implementation example:

[0073] Considering the correlation between data of different dimensions, the historical data are recorded in multiple time periods. With the Correlation between dimension data for:

[0074]

[0075] in, It represents the function for finding the correlation coefficient between two sequences.

[0076] Then, according to the correlation between the two dimensional data, the weight of each dimension is obtained. When there is a high collinearity between a single dimension and the other multiple dimensions, it is more likely to lead to redundant data fusion when making actual judgments. The weight of the dimension for:

[0077]

[0078] in, Represents the normalization function.

[0079] Step S150: Obtaining a fault prediction result of the diesel engine based on the fused BPA function.

[0080] Preferably, in one embodiment of the present invention, step S150 may include: obtaining the support level of the target window data for each type of fault based on the fused BPA function; and determining that the diesel engine has experienced a corresponding type of fault when the support level for the corresponding type of fault exceeds a preset decision threshold. For example, after completing the multi-dimensional information fusion and obtaining the final BPA function, each type of fault (such as fuel system fault, cooling system abnormality, intake system fault, etc.) is assigned a specific confidence value, indicating the support level of the current system state for that fault. The higher the confidence level, the closer the system characteristics are to the characteristic pattern of that type of fault in historical data. Therefore, the BPA function can be viewed as a distribution of matching degrees between the current diesel engine state and various potential fault types.

[0081] It should be noted that in practical applications, a decision threshold is typically set to ensure a timely and reliable fault warning mechanism. For example, if the confidence level of a particular fault type exceeds 0.6 (i.e., a confidence level greater than 60%), the system can preliminarily determine that the diesel engine has a high probability of experiencing this type of fault and trigger a corresponding warning response or prompt the operator to conduct further diagnostic inspections. This judgment method based on maximum confidence balances the credibility of the fusion results with the timeliness of the warning, avoiding false positives or missed negatives caused by excessive uncertainty.

[0082] So far, the present invention is completed.

[0083] In summary, in an embodiment of the present invention, a historical fault sample set and target window data of a diesel engine are obtained; based on the historical fault sample set, fault associations between historical operating status data of each dimension and various types of faults are obtained; based on the data differences between the historical operating status data of each dimension and the target window data, a single-dimensional BPA function is obtained; based on the correlations between the historical operating status data of each dimension, the single-dimensional BPA functions are fused to obtain a fused BPA function; and based on the fused BPA function, a fault prediction result for the diesel engine is obtained. By introducing the Dempster-Shafer evidence theory, the present invention fuses the multi-dimensional data collected during diesel engine operation, fully utilizing the responsiveness of various sensors to fault characteristics under different operating conditions. By evaluating the representation strength of each dimension and each type of fault in the historical data, a dynamic BPA function is constructed, thereby enhancing the pertinence and sensitivity of the early warning. Furthermore, by improving the fusion mechanism and introducing a correlation correction strategy between data dimensions, the impact of redundant interference on the diagnostic results is effectively reduced. The above-mentioned diesel engine fault warning method based on multi-source data fusion significantly improves the accuracy and robustness of the diesel engine fault warning system while ensuring the algorithm's interpretability and real-time performance. It can be widely used in intelligent operation and maintenance and industrial health management scenarios.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A diesel engine fault early warning method based on multi-source data fusion, characterized in that: The method comprises: Obtaining a historical fault sample set and target window data of the diesel engine; wherein the target window data includes operating status data at the current moment and operating status data of a preset time period selected from the current moment; the historical fault sample set and the target window data are both multidimensional data; Based on the historical fault sample set, obtaining the fault correlation between the historical operating status data of each dimension and each type of fault; Obtaining a single-dimensional BPA function based on the data difference between the historical operating status data of each dimension and the target window data; Based on the correlation between the historical operating status data of each dimension, the single-dimensional BPA functions are fused to obtain a fused BPA function; Obtaining a fault prediction result of the diesel engine based on the fused BPA function; The obtaining of a single-dimensional BPA function based on the data difference between the historical operating status data of each dimension and the target window data includes: obtaining a direct support degree of the target window data of each dimension for each type of fault based on the dynamic time warping (DTW) distance between the historical operating status data of each dimension and the target window data; and correcting the direct support degree of the target window data of each dimension for each type of fault based on the fault correlation to obtain a single-dimensional BPA function. The single-dimensional BPA functions are fused based on the correlation between the historical operating status data of each dimension to obtain a fused BPA function, including: fusing the single-dimensional BPA functions of two different dimensions to obtain an initial fused BPA function; and correcting the initial fused BPA function based on the conflict degree of the two different dimensions to obtain a fused BPA function; wherein the conflict degree is used to characterize the degree of conflict between the single-dimensional BPA functions of different dimensions when supporting different types of faults.

2. The diesel engine fault early warning method based on multi-source data fusion according to claim 1 is characterized in that: The acquisition of the fault correlation between the historical operating status data of each dimension and each type of fault includes: Obtain the Pearson correlation coefficient between the historical operating status data of each dimension and each type of fault.

3. The diesel engine fault early warning method based on multi-source data fusion according to claim 2 is characterized in that: The obtaining of the fault correlation between the historical operating status data of each dimension and each type of fault further includes: For the same type of fault, a fluctuation penalty term is obtained based on the standard deviation of the historical operating status data of each dimension under the same fault severity; wherein the fault severity includes fault not occurring, fault incipient, fault aggravation, and fault causing equipment failure; The Pearson correlation coefficient is corrected based on the fluctuation penalty term to obtain the fault correlation between the historical operating status data of each dimension and each type of fault.

4. The diesel engine fault early warning method based on multi-source data fusion according to claim 1 is characterized in that: After obtaining the single-dimensional BPA function, the method further includes: The single-dimensional BPA function is normalized.

5. The diesel engine fault early warning method based on multi-source data fusion according to claim 1 is characterized in that: The fusing of the single-dimensional BPA functions of two different dimensions to obtain an initial fused BPA function includes: Obtaining a fusion weight for each dimension; wherein the fusion weight is used to characterize the importance of the single-dimensional BPA function of the corresponding dimension when fused with the single-dimensional BPA function of other dimensions; Based on the fusion weight, the single-dimensional BPA functions of two different dimensions are fused to obtain an initial fused BPA function.

6. The diesel engine fault early warning method based on multi-source data fusion according to claim 5 is characterized in that: The initial fusion BPA function is modified based on the conflict degrees of two different dimensions to obtain a fusion BPA function, including: Based on the fusion weight of each dimension, the conflict degree of two different dimensions is corrected to obtain a corrected conflict degree; Based on the modified conflict degree, the initial fused BPA function is modified to obtain a fused BPA function.

7. The diesel engine fault early warning method based on multi-source data fusion according to claim 1 is characterized in that: The obtaining of the fusion weight of each dimension includes: Obtaining a negative value of a correlation between the historical operating status data of each dimension and the historical operating status data of other dimensions, and performing normalization processing on the negative value of the correlation; For each dimension, the negative values ​​of the normalized correlations between the current dimension and other dimensions are averaged to obtain the fusion weight of each dimension.

8. The diesel engine fault early warning method based on multi-source data fusion according to claim 1 is characterized in that: The obtaining of the fault prediction result of the diesel engine based on the fused BPA function includes: Based on the fused BPA function, obtaining the support degree of the target window data for each type of fault; When the support level of the corresponding type of fault is greater than a preset decision threshold, it is determined that the diesel engine has a corresponding type of fault.

Images