Marine diesel engine intelligent fault diagnosis method based on interval belief rule base

Abstract

The invention discloses a marine diesel engine intelligent fault diagnosis method based on an interval belief rule base, and belongs to the technical field of fault diagnosis, and the method comprises the following steps: 1, obtaining an actual measurement data sample of a marine diesel engine; the method comprises the following steps: under different fault conditions, selecting a proper acquisition position in a marine engine, and selecting a proper sensor to be placed at a specified position according to a test requirement; according to the method, intelligent fault diagnosis is carried out through the interval belief rule base (IBRB), the uncertainty in operation data of the diesel engine is fully considered, the system can still keep efficient and stable diagnosis capacity under the complex environment and the ocean working condition through robustness analysis of the model, the reliability and practicability of the system are further improved, and therefore the method has the advantages of being high in practicability and high in practicability. The diagnosis method provided by the invention has higher accuracy, robustness and real-time performance, can adapt to a complex marine environment, and provides more reliable technical support for fault diagnosis of the diesel engine.

Description

Technical Field

[0001] The invention belongs to the technical field of fault diagnosis, and in particular relates to an intelligent fault diagnosis method for a marine diesel engine based on an interval confidence rule base. Background Art

[0002] Marine diesel engines, as the main power source of ships, play a vital role. Diesel engines work in complex and changeable marine environments for a long time and are easily affected by external environmental factors (such as high temperature, high humidity, corrosion, etc.), resulting in unstable working conditions or failures. Failure of marine diesel engines will not only cause the ship to stop operating, but in serious cases will also cause safety accidents, endangering personal and property safety. Therefore, ensuring the reliability and stability of diesel engines is crucial to the safe operation of the entire ship system. Summary of the invention

[0003] The purpose of the present invention is to solve the problems raised in the above background technology and to propose an intelligent fault diagnosis method for marine diesel engines based on an interval confidence rule base.

[0004] In order to achieve the above object, the present invention adopts the following technical scheme: a marine diesel engine intelligent fault diagnosis method based on interval confidence rule base, comprising the following steps:

[0005] Step 1: Obtain measured data samples of marine diesel engines;

[0006] By selecting the appropriate collection position on the marine engine under different fault conditions, selecting the appropriate sensor and placing it at the specified position according to the test needs, the vibration signal emitted by the engine is collected, sorted and the measured data samples are obtained;

[0007] Step 2: Construct an indicator system for fault diagnosis;

[0008] Feature extraction is performed on the measured data set obtained in step 1, and the multi-dimensional features of the vibration signal such as time domain and frequency domain are combined for analysis. Based on the effectiveness and wide application of these features in mechanical fault diagnosis, domain experts select key features that have an important impact on the health status of marine diesel engines, and use them to construct an indicator system for fault diagnosis;

[0009] Step 3: Build an expert knowledge base and indicator reference interval;

[0010] Experts in related fields build an expert knowledge base based on the analysis of marine diesel engine fault data and their understanding of its principles, and determine the reference range of the indicators based on the indicator system obtained in step 2;

[0011] Step 4: Construct an inference model based on interval confidence rule base;

[0012] Reasoning is performed based on the interval confidence rule base. First, the input data is matched with the reference interval of each rule, and the corresponding rules are activated by calculating the matching degree of the input data in the reference interval. Then, the evidential reasoning (ER) method is used to integrate these activated rules and finally output the fault diagnosis results of the marine diesel engine.

[0013] Step 5: Calculate the reliability of expert knowledge;

[0014] Calculate the reliability of expert knowledge. This step evaluates the reliability of the initial model parameters constructed by expert knowledge to ensure that the credibility of expert knowledge is reasonably introduced into the model, thereby providing a reliable knowledge basis for subsequent reasoning.

[0015] Step 6: Optimize the model and perform interpretability analysis;

[0016] The model is optimized based on the reliability of the expert knowledge calculated in step 5. This step reasonably introduces the credibility of expert knowledge in the optimization process to ensure that the model not only relies on data during optimization, but also balances expert knowledge, thereby enhancing the interpretability of the model and further improving the accuracy and robustness of the diagnostic results.

[0017] Step 7: Consider indicator uncertainty processing;

[0018] Dealing with uncertainty in input data, by introducing a calculation method for attribute reliability, evaluating data reliability based on the fluctuation and noise of sensor data, and comprehensively considering data uncertainty in the reasoning process;

[0019] Step 8: Robustness analysis of the model;

[0020] Based on the results of uncertainty processing in step 7, further robustness analysis is performed to evaluate the stability of the model in the face of external disturbances such as noise and data fluctuations, ensuring that the model can still maintain accuracy and robustness in multi-source data and complex environments. Combined with the evaluation of attribute reliability, the robustness of fault diagnosis of the system under different working conditions is verified;

[0021] Step 9: Obtaining the fault diagnosis result of the marine diesel engine;

[0022] Obtain the fault diagnosis results of the marine diesel engine; finally output the fault diagnosis results of the marine diesel engine by integrating the multi-source input data, expert knowledge and optimized reasoning of the model processed in the above steps. This step is based on the reasoning process of the interval confidence rule base, combined with the analysis of uncertainty and robustness, to ensure that the diagnosis results have high accuracy and stability in complex working environments.

[0023] As a further description of the above technical solution: the types of vibration signal data samples obtained in step 1 include but are not limited to acceleration signals, velocity signals, displacement signals and vibration frequency spectra.

[0024] As a further description of the above technical solution: the construction of the fault diagnosis index system in the step 2 also includes processing the measured data obtained in the step 1. The processing flow first pre-processes the vibration signal to remove noise and interference signals to ensure the accuracy and stability of the data. Next, feature extraction is performed, and multi-dimensional analysis is performed in combination with time domain features, frequency domain features and time-frequency domain features: time domain features such as mean, variance, kurtosis and skewness are used to measure the basic statistical characteristics of the vibration signal; frequency domain features analyze the main frequency, spectral energy distribution and power spectral density of the vibration signal through frequency components to help identify the contribution of different frequency components to the fault; time-frequency domain features capture the changes in signal time and frequency through methods such as wavelet transform and short-time Fourier transform. After feature extraction, these features are screened using the knowledge of domain experts, and key features closely related to the health status of marine diesel engines are selected to construct a fault diagnosis index system. This index system will include a series of numerical features that reflect different aspects of the bearing health status. These indicators will become the input for constructing an interval confidence rule base fault diagnosis model in subsequent steps.

[0025] As a further description of the above technical solution: in the step three, an expert knowledge base and an indicator reference interval are constructed. First, based on the experience of experts in the field and historical fault data, key fault influencing factors in the operation of marine diesel engines are identified, and an expert knowledge base is constructed. The knowledge base includes information about various fault modes, operating states and corresponding diagnostic characteristics; then, in the process of dividing the reference interval, each diagnostic indicator is divided into multiple reference intervals according to the fluctuation range of the actual measured data and the statistical characteristics of the historical data; each reference interval represents the range of variation of a specific indicator under different health states, which can effectively cope with the fluctuation and uncertainty of the data during operation; experts refine and adjust the upper and lower limits of these intervals based on their experience to ensure that each reference interval can accurately reflect the characteristic changes under different fault modes.

[0026] As a further description of the above technical solution: in the step 4, the inference model based on the interval confidence rule base is constructed. First, based on the initial parameters for constructing the interval confidence rule base that can be obtained in step 3, the Kth rule of the IF-THEN rule based on the interval confidence rule base is expressed as:

[0027]

[0028] where x i(i=1,…,M) represents the premise attribute fault feature in the input model, M is the number of premise attributes, [a i ,b i ] represents the reference interval set of the i-th indicator, where i=1,...,Mr k represents the reliability of the rule; w k represents the rule weight, L is the number of rules, β N,k Indicates D N The confidence level is the Nth failure mode, where N represents the number of failure modes.

[0029] When a set of features (x1, x2, ..., x M )In the input model, each feature x i (1≤i≤M) falls within an interval, activating a rule. For feature x i (1≤i≤M), the rule matching degree calculation formula is:

[0030]

[0031] where a i represents the interval rule matching degree corresponding to the i-th attribute, Mid = (a i +b i ) / 2 represents the position of the midpoint of the interval;

[0032] After calculating the rule matching degree, the i-th feature x i The activation weight of the (1≤i≤M) activated rule is calculated as follows:

[0033] w i =[ε+(1-ε)×]×λ i (3)

[0034] Among them, the activation weight of the activated i-th interval rule is recorded as w i , initial rule weight λ i ; The minimum activation weight ε is given by expert knowledge and can be any constant in the range of 0 to 1;

[0035] ER rules provide a transparent and intuitive framework for evidence combination. In addition, expert knowledge can be integrated, and the impact of different evidence on the entire reasoning process can be easily explained and evaluated. ER rules provide a reliable solution for decision-making and evaluation with multiple evidences.

[0036] The rules in the model building process are actually used as evidence in the ER rules. L independent pieces of evidence are recorded as e i (i=1,...,L), the identification framework is denoted as Θ, which consists of N evaluation levels D n(n=1,...,N), which can be expressed as Θ={D1,...,D N}, then a piece of evidence can be represented as the following confidence distribution:

[0037]

[0038] Among them, β n,i is represented as the evaluation scheme in evidence e i The following is evaluated as D n The confidence level, β Θ,i is represented as global ignorance, i.e., the confidence of the i-th premise attribute relative to the identification framework Θ;

[0039] Assume that the weight of evidence is w i (i=1,...,L), and w i ∈[0,1]; the reliability of evidence is r i (i=1,...,L), and r i ∈[0,1]; then, the confidence distribution with mixed weighting of reliable evidence can be expressed as:

[0040]

[0041] Among them, the power set is denoted as β(Θ), and the i-th attribute is at level D n The probability mass of the mixture under It can be obtained by the following formula:

[0042]

[0043] Among them, the normalization coefficient is denoted as c rw,i =1 / (1+w i -r i ), which satisfies The joint support of any two pieces of evidence is β n,e(2) The calculation is as follows:

[0044]

[0045] Then, the joint support of L independent pieces of evidence β n,e(L) It can be calculated in the following general way:

[0046]

[0047] Where k = 3, 4, ..., L, β n,e(k) It is the first k attributes fused relative to level D n The confidence level, and m n,e(1) =m n,1 ,m β(Θ),e(1) =mβ(Θ),1 , calculated by the above formula, the comprehensive evaluation results can be obtained as follows:

[0048] e(L)={(D n ,β n,e(L) ),n=1,...,N,(Θ,β Θ,e(L) )} (13)

[0049] Will be graded D n The utility of n ), then we can get the final output result y, that is, the expected utility value. The method for calculating the final expected utility value is as follows:

[0050]

[0051] As a further description of the above technical solution: In step 5, the reliability of expert knowledge is calculated: Specifically, the reliability of expert knowledge is not only based on the initial parameters set by the expert, but also combined with objective analysis to quantify the credibility of expert knowledge. First, assuming that the observed data of the i-th attribute is x i (n)(n=1,2,...,N,i=1,2,...,M), then, the expert system constructed by expert knowledge constructs an interval confidence rule base to statistically evaluate the initial parameters and obtain the reliability evaluation value, the evaluation value r e The calculation of is based on comparing the predicted result with the actual state, using the mean squared error to quantify this difference:

[0052]

[0053] where k is the credibility of expert knowledge assessed through historical records and peer review, and r e =0 means that the expert knowledge is completely reliable.

[0054] As a further description of the above technical solution: Step 6 optimizes the model and performs interpretability analysis:

[0055] The P-CMA-ES method is used to optimize the confidence, rule weight and rule reliability in the initial rule base to improve the accuracy of the state assessment model. Since the optimization process will have a negative impact on the interpretability of the method, interpretability constraints are added during the optimization process so that the optimization results do not violate the initial expert judgment and each optimized parameter maintains its original physical meaning. To build an optimization model, we must first clarify the function to be optimized. The function with interpretable optimization is expressed as:

[0056]

[0057] Among them, output modeland output actual Respectively represent the predicted value and true value of the interval confidence rule base, T represents the number of data samples, (β, λ, r) initial represents the initial parameter values ​​set by expert knowledge, (β,λ,r) optimal represents the optimized parameters, (β,λ,r) low and (β,λ,r) up Represents the variation space of the parameters given by the expert;

[0058] A reasonable and interpretable confidence distribution can be either monotonic or convex, but not concave:

[0059]

[0060] As a further description of the above technical solution: the step 7 considers the indicator uncertainty processing. In engineering practice, the observation data may be interfered with, resulting in the observation data being unable to correctly reflect the system information, thereby causing the data to be unreliable. The attribute reliability is calculated by setting a tolerance range to eliminate the interference of uncertain data;

[0061] Let all inputs corresponding to the i-th attribute be x i,1 ,x i,2 ,...,x i,T ,i=1,2...,M means,x i,1 ,x i,2 ,...,x i,T The mean of The corresponding standard deviation is σ i ,i=1,2,...,M, so the tolerance range is Where λ is the adjustment factor of the tolerance range, which is usually determined based on expert knowledge. or When ij =1 means the observation is unreliable, otherwise u ij =0 means the observation is reliable. represents the number of unreliable observations, so the reliability of the i-th attribute is expressed as:

[0062]

[0063] Among them, the reliability of the i-th attribute is represented by h i (0≤h i ≤1) indicates that h i =1 means the attribute is completely reliable, h i =0 means completely unreliable.

[0064] As a further description of the above technical solution: After the uncertainty is processed in step 7, the robustness of the model is further analyzed in step 8. In order to ensure the stability and reliability of the model under different input conditions, the Lipschitz condition is used to measure the disturbance amplitude and robustness of the model. Specifically, the reaction degree of the model to the input disturbance is calculated based on the Hatton distance to ensure that the model can still remain robust in a complex environment. The Lipschitz condition of the interval confidence rule base is defined as follows

[0065] for The Interval Confidence Rule Model (IBRB) satisfies If there is a very small constant Make all

[0066] This model is called Lipschitz stable, where d(·) is the distance metric, The disturbance The perturbation of x is represented by x′, and the Lipschitz condition at the Manhattan distance can be rewritten as:

[0067]

[0068] The modulo operation is represented by |·|

[0069] The above definition gives the calculation of the Lipschitz constant for IBRB. Furthermore, for There is a functional relationship: If part of its function is As shown below:

[0070]

[0071] where x′ is the perturbation of x, c i is a small constant. When c i When ≤θ, we say that the function f is Lipschitz stable, and the corresponding Lipschitz constant is θ, where θ can be regarded as a perturbation constant. The smaller the constant, the more stable the model and the better the robustness.

[0072] Taking the limit on both sides of equation (19) yields:

[0073]

[0074] At this time, the complex calculation of the Lipschitz constant is transformed into the calculation of the partial derivative of the high-dimensional function. By calculating the partial derivative of the high-dimensional function, the Lipschitz constant of the function relationship f can be calculated, and then the perturbation constant θ that makes the function f stable can be obtained, thereby realizing the robustness measurement of the function.

[0075] As a further description of the above technical solution: Step nine obtains the fault diagnosis result of the marine diesel engine. After completing the reasoning, optimization and robustness analysis of the previous steps, the fault diagnosis result of the marine diesel engine is obtained. Through the comprehensive reasoning of input data processing, application of expert knowledge base, uncertainty processing and robustness analysis, the model finally outputs the engine fault diagnosis result. This result is not only based on the fusion analysis of multi-dimensional data, but also takes into account the uncertainty in the diagnosis process and the impact of complex environment on the data, ensuring the high accuracy and reliability of the fault diagnosis result.

[0076] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0077] 1. The present invention uses an interval confidence rule base (IBRB) to perform intelligent fault diagnosis, fully considering the uncertainty in the diesel engine operation data, and uses interval form to represent various attribute data, so that the diagnosis results are more robust and accurate. Secondly, combined with the construction of the expert knowledge base and reliability calculation, the expert experience and data analysis are organically integrated, which greatly improves the interpretability and credibility of the diagnosis model. When processing multi-source data, the model effectively integrates various types of information through the evidence reasoning (ER) method, resolves the conflicts between different data sources, and ensures the intelligence and comprehensiveness of the fault diagnosis process. In addition, the model uses a tolerance range to process the uncertainty of the indicators, effectively eliminating the influence of unreliable data. Finally, after the robustness analysis of the model, it is ensured that the system can still maintain efficient and stable diagnostic capabilities in complex environments and marine conditions, further improving the reliability and practicality of the system. Therefore, the diagnostic method of the present invention has higher accuracy, robustness and real-time performance, can adapt to complex marine environments, and provide more reliable technical support for diesel engine fault diagnosis.

[0078] 2. The present invention combines expert knowledge and collected measured data samples to extract key factors affecting the health status of marine diesel engines, construct a fault diagnosis index system, and construct an expert knowledge base. In combination with expert knowledge, an inference model based on an interval confidence rule base is constructed. Multi-source data is processed and analyzed through interval-based attributes and confidence distributions. This method uses a tolerance range to eliminate unreliable data and ensure the stability and accuracy of input data. In the reasoning process, the method combines the evidence reasoning method to gradually integrate various types of data and rules and output preliminary fault diagnosis results. Subsequently, the system enhances the adaptability of the model in complex environments by optimizing and analyzing the robustness of the preliminary results, and ultimately obtains comprehensive fault diagnosis results.

[0079] 3. The present invention constructs a reliable fault diagnosis model by combining expert knowledge and data analysis. The model realizes data fusion and uncertainty processing by combining the interval confidence rule base with the tolerance range, making the diagnosis result more accurate and reliable. In addition, the evidence reasoning method effectively solves the conflict between multi-source data, further improves the comprehensiveness and intelligence level of diagnosis, and enhances the interpretability and robustness of the diagnosis model through the optimization process, ensuring that the parameters are in line with the range of actual working conditions. The present invention realizes the accuracy and reliability of marine diesel engine fault diagnosis through the above method, and the diagnosis process is more transparent and suitable for complex marine environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 A step diagram of the intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base proposed by the present invention;

[0081] Figure 2 A method flow chart of the intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base proposed by the present invention;

[0082] Figure 3 A schematic diagram of the method reasoning process of the marine diesel engine intelligent fault diagnosis method based on interval confidence rule base proposed by the present invention;

[0083] Figure 4 A schematic diagram of the reliability calculation process of the expert knowledge of the method for intelligent fault diagnosis of marine diesel engines based on interval confidence rule base proposed by the present invention;

[0084] Figure 5 It is a schematic diagram of the process of considering the uncertainty of indicators in the intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base proposed by the present invention;

[0085] Figure 6 This is a schematic diagram of the robustness analysis of the intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base proposed by the present invention. DETAILED DESCRIPTION

[0086] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0087] Please refer to the attached Figure 1 -Attached Figure 6The present invention provides a technical solution: an intelligent fault diagnosis method for a marine diesel engine based on an interval confidence rule base, comprising the following steps:

[0088] Step 1: Obtain measured data samples of marine diesel engines;

[0089] Step 2: Construct an indicator system for fault diagnosis;

[0090] Step 3: Build an expert knowledge base and indicator reference interval;

[0091] Step 4: Construct an inference model based on interval confidence rule base;

[0092] Step 5: Calculate the reliability of expert knowledge;

[0093] Step 6: Optimize the model and perform interpretability analysis;

[0094] Step 7: Consider indicator uncertainty processing;

[0095] Step 8: Robustness analysis of the model;

[0096] Step 9: Obtain the fault diagnosis results of the marine diesel engine.

[0097] In the above step 1, by selecting a suitable acquisition position on the marine engine under different fault conditions, a suitable sensor is selected and placed at a specified position according to the test needs, the vibration signal emitted by the engine is collected, sorted and measured data samples are obtained; the types of vibration signal data samples obtained include but are not limited to acceleration signals, velocity signals, displacement signals and vibration frequency spectra.

[0098] In the above step 2, feature extraction is performed on the measured data set obtained in step 1, and the multi-dimensional features such as time domain and frequency domain of the vibration signal are combined for analysis. Based on the effectiveness and wide application of these features in mechanical fault diagnosis, field experts select key features that have an important impact on the health status of marine diesel engines, and use them to construct an indicator system for fault diagnosis. Specifically, it includes processing and feature extraction of the measured data obtained in step 1. The processing flow first pre-processes the vibration signal to remove noise and interference signals to ensure the accuracy and stability of the data. Next, feature extraction is performed, and multi-dimensional analysis is performed in combination with time domain features, frequency domain features, and time-frequency domain features: time domain features such as mean, variance, kurtosis, and skewness are used to measure the basic statistical characteristics of vibration signals; frequency domain features analyze the main frequency, spectral energy distribution, and power spectral density of vibration signals through frequency components to help identify the contribution of different frequency components to faults; time-frequency domain features capture the changes in signal time and frequency through methods such as wavelet transform and short-time Fourier transform. After feature extraction, the knowledge of domain experts is used to screen these features, select key features that are closely related to the health of marine diesel engines, and build an indicator system for fault diagnosis. This indicator system will include a series of numerical features that reflect different aspects of the bearing health. These indicators will become the input for building the interval confidence rule base fault diagnosis model in the subsequent steps.

[0099] In the above step three, experts in related fields build an expert knowledge base based on their understanding of the fault data analysis principle of marine diesel engines. And according to the indicator system obtained in step two, determine the reference range of the indicator. First, based on the experience of experts in the field and historical fault data, identify the key fault influencing factors in the operation of marine diesel engines and build an expert knowledge base. The knowledge base includes information about various fault modes, operating states and corresponding diagnostic features. Subsequently, in the process of dividing the reference interval, each diagnostic indicator is divided into multiple reference intervals according to the fluctuation range of the actual measured data and the statistical characteristics of the historical data. Each reference interval represents the range of variation of a specific indicator under different health states, which can effectively deal with the fluctuation and uncertainty of data during operation. Experts refine and adjust the upper and lower limits of these intervals based on experience to ensure that each reference interval can accurately reflect the characteristic changes under different fault modes.

[0100] In the above step 4, reasoning is performed based on the interval confidence rule base. First, the input data is matched with the reference interval of each rule, and the corresponding rule is activated by calculating the matching degree of the input data in the reference interval. Then, the evidence reasoning (ER) method is used to integrate these activated rules. Finally, the fault diagnosis result of the marine diesel engine is output. First, based on the initial parameters for constructing the interval confidence rule base that can be obtained in step 3, the Kth rule of the IF-THEN rule based on the interval confidence rule base is expressed as:

[0101]

[0102] where x i (i=1,…,M) represents the premise attribute fault feature in the input model. M is the number of premise attributes. i ,b i ] represents the reference interval set of the i-th indicator. Where i=1,…,Mr k Indicates the reliability of the rule. k represents the rule weight. L is the number of rules, β N,k Indicates D N The confidence level is , this is the Nth failure mode. N is the number of failure modes.

[0103] When a set of features (x1, x2, ..., x M )In the input model, each feature x i (1≤i≤M) falls within an interval, activating a rule. i (1≤i≤M), the rule matching degree calculation formula is:

[0104]

[0105] where a i Indicates the matching degree of the interval rule corresponding to the i-th attribute. Mid=(a i +b i ) / 2 represents the position of the midpoint of the interval.

[0106] After calculating the rule matching degree, the i-th feature x i The activation weight of the (1≤i≤M) activated rule is calculated as follows:

[0107] w i =[ε+(1-ε)×]×λ i (3)

[0108] Among them, the activation weight of the activated i-th interval rule is recorded as w i , initial rule weight λ iThe minimum activation weight ε is given by expert knowledge and can be any constant in the range of 0 to 1.

[0109] ER rules provide a transparent and intuitive framework for evidence combination. In addition, expert knowledge can be integrated, and the impact of different evidences on the entire reasoning process can be easily explained and evaluated. ER rules provide a reliable solution for decision-making and evaluation with multiple evidences.

[0110] The rules in the model building process are actually used as evidence in the ER rules. Let L independent pieces of evidence be denoted as e i (i=1,...,L). The identification framework is denoted as Θ, which consists of N evaluation levels D n (n=1,...,N). This can be expressed as Θ={D1,…,D N}. Then, a piece of evidence can be represented as the following confidence distribution:

[0111]

[0112] Among them, β n,i is represented as the evaluation scheme in evidence e i The following is evaluated as D n Confidence level. Θ,i is represented as global ignorance, i.e., the confidence of the i-th premise attribute relative to the identification framework Θ.

[0113] Assume that the weight of evidence is w i (i=1,…,L), and w i ∈[0,1]; the reliability of evidence is r i (i=1,…,L), and r i ∈[0,1]. Then, the confidence distribution with weighted evidence mixture with reliability can be expressed as:

[0114]

[0115] The power set is denoted as β(Θ). The i-th attribute is at level D n The probability mass of the mixture under It can be obtained by the following formula:

[0116]

[0117] Among them, the normalization coefficient is denoted as c rw,i =1 / (1+w i -r i ), which satisfies The joint support of any two pieces of evidence is β n,e(2) The calculation is as follows:

[0118]

[0119] Then, the joint support of L independent pieces of evidence β n,e(L) It can be calculated in the following general way:

[0120]

[0121] Where k = 3, 4, ..., L. β n,e(k) It is the first k attributes after fusion relative to level D n The confidence level, and m n,e(1) =m n,1 ,m β(Θ),e(1) =m β(Θ),1 . Through the above formula calculation, the comprehensive evaluation results can be obtained as follows:

[0122] e(L)={(D n ,β n,e(L) ),n=1,...,N,(Θ,β Θ,e(L) )} (13)

[0123] Will be graded D n The utility of n ). Then we can get the final output result y, which is the expected utility value. The method to calculate the final expected utility value is as follows:

[0124]

[0125] In the above step 5, the reliability of expert knowledge is calculated. This step evaluates the reliability of the initial model parameters constructed by expert knowledge to ensure that the credibility of expert knowledge is reasonably introduced into the model, thereby providing a reliable knowledge basis for subsequent reasoning. Specifically, the reliability of expert knowledge is not only based on the initial parameters set by the expert, but also combines objective analysis to quantify the credibility of expert knowledge. First, assume that the observed data of the i-th attribute is x i (n)(n=1,2,…,N,i=1,2,…,M). Then, the expert system constructed by expert knowledge is used to statistically evaluate the initial parameters and obtain the reliability evaluation value. e The calculation of is based on comparing the predicted result with the actual state, using the mean squared error to quantify this difference:

[0126] where k is the credibility of the expert’s knowledge as assessed by historical records and peer review. e =0 means that the expert knowledge is completely reliable.

[0127] In the above step 6, the model is optimized based on the reliability of the expert knowledge calculated in step 5. This step reasonably introduces the credibility of expert knowledge in the optimization process to ensure that the model not only relies on data during optimization, but also balances expert knowledge, thereby enhancing the interpretability of the model and further improving the accuracy and robustness of the diagnostic results.

[0128] The P-CMA-ES method is used to optimize the confidence, rule weight and rule reliability in the initial rule base to improve the accuracy of the state assessment model. Since the optimization process will have a negative impact on the interpretability of the method, interpretability constraints are added during the optimization process so that the optimization results do not violate the initial expert judgment and each optimized parameter maintains its original physical meaning. To build an optimization model, we must first clarify the function to be optimized. The function with interpretable optimization is expressed as:

[0129]

[0130] Among them, output model and output actual Respectively represent the predicted value and true value of the interval confidence rule base, and T represents the number of data samples. (β,λ,r) initial represents the initial parameter values ​​set by expert knowledge. (β,λ,r) optimal represents the optimized parameters. (β,λ,r) low and (β,λ,r) up Represents the variation space of the parameters given by the expert.

[0131] A reasonable and interpretable confidence distribution can be either monotonic or convex, but not concave:

[0132]

[0133] In the above step 7, the uncertainty in the input data is processed. By introducing the calculation method of attribute reliability, the data reliability is evaluated based on the fluctuation and noise of sensor data, and the uncertainty of data is comprehensively considered in the reasoning process. Consider the uncertainty processing of indicators. In engineering practice, the observed data may be interfered with, resulting in the observed data not correctly reflecting the system information, thus causing the data to be unreliable. The attribute reliability is calculated by setting the tolerance range to eliminate the interference of uncertain data.

[0134] Let all inputs corresponding to the i-th attribute be x i,1 ,x i,2 ,…,x i,T ,i=1,2…,M represents. x i,1 ,x i,2 ,...,x i,T The mean of The corresponding standard deviation is σ i ,i=1,2,...,M. Therefore, the tolerance range is Where λ is the adjustment factor of the tolerance range, which is usually determined based on expert knowledge. or When ij =1 indicates that the observed value is unreliable. Otherwise u ij =0 means the observation is reliable. represents the number of unreliable observations. Therefore, the reliability of the i-th attribute is expressed as:

[0135]

[0136] Among them, the reliability of the i-th attribute is represented by h i (0≤h i ≤1) indicates that h i =1 means the attribute is completely reliable, h i =0 means completely unreliable.

[0137] Based on the results of the uncertainty processing in step seven in step eight above, a robustness analysis is further performed. By evaluating the stability of the model in the face of external disturbances such as noise and data fluctuations, it is ensured that the model can still maintain accuracy and robustness in multi-source data and complex environments. Combined with the evaluation of attribute reliability, the fault diagnosis robustness of the system under different working conditions is verified. The robustness of the model is further analyzed. In order to ensure the stability and reliability of the model under different input conditions, the Lipschitz condition is used to measure the disturbance amplitude and robustness of the model. Specifically, the response degree of the model to the input disturbance is calculated based on the Haddon distance to ensure that the model can remain robust in a complex environment. The Lipschitz condition that defines the interval confidence rule base is as follows

[0138] for The Interval Confidence Rule Model (IBRB) satisfies If there is a very small constant Make all

[0139] This model is called Lipschitz stable, where d(·) is the distance metric, The disturbance The perturbation of x is represented by x′. The Lipschitz condition at the Manhattan distance can be rewritten as:

[0140]

[0141] The modulo operation is represented by |·|

[0142] The above definition gives the calculation of the Lipschitz constant for IBRB. Furthermore, for There is a functional relationship: If part of its function is As shown below:

[0143]

[0144] where x′ is the perturbation of x. i is a small constant. i When ≤θ, we say that the function f is Lipschitz stable, and the corresponding Lipschitz constant is θ. Among them, θ can be regarded as a perturbation constant. The smaller the constant, the more stable the model and the better the robustness.

[0145] Taking the limit on both sides of equation (19) yields:

[0146]

[0147] At this point, the complex calculation of the Lipschitz constant is transformed into the calculation of the partial derivative of the high-dimensional function. By calculating the partial derivative of the high-dimensional function, the Lipschitz constant of the function relationship f can be calculated, and then the perturbation constant θ that stabilizes the function f can be obtained, thereby realizing the robustness measurement of the function.

[0148] Based on the above step nine, the fault diagnosis results of the marine diesel engine are obtained. By integrating the multi-source input data, expert knowledge and optimized reasoning of the model processed in the above steps, the fault diagnosis results of the marine diesel engine are finally output. This step is based on the reasoning process of the interval confidence rule base, combined with the analysis of uncertainty and robustness, to ensure that the diagnosis results have high accuracy and stability in complex working environments. After completing the reasoning, optimization and robustness analysis of the previous steps, the fault diagnosis results of the marine diesel engine are obtained. Through the comprehensive reasoning of the processing of input data, the application of expert knowledge base, uncertainty processing and robustness analysis, the model finally outputs the fault diagnosis results of the engine. This result is not only based on the fusion analysis of multi-dimensional data, but also takes into account the uncertainty in the diagnosis process and the impact of complex environment on data, ensuring the high accuracy and reliability of the fault diagnosis results.

[0149] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base, characterized in that: The following steps are involved: Step 1: Obtain measured data samples of marine diesel engines; By selecting the appropriate collection position on the marine engine under different fault conditions, selecting the appropriate sensor and placing it at the specified position according to the test needs, the vibration signal emitted by the engine is collected, sorted and the measured data samples are obtained; Step 2: Construct an indicator system for fault diagnosis; Feature extraction is performed on the measured data set obtained in step 1, and the multi-dimensional features of the vibration signal such as time domain and frequency domain are combined for analysis. Based on the effectiveness and wide application of these features in mechanical fault diagnosis, domain experts select key features that have an important impact on the health status of marine diesel engines, and use them to construct an indicator system for fault diagnosis; Step 3: Build an expert knowledge base and indicator reference interval; Experts in related fields build an expert knowledge base based on the analysis of marine diesel engine fault data and their understanding of its principles, and determine the reference range of the indicators based on the indicator system obtained in step 2; Step 4: Construct an inference model based on interval confidence rule base; Reasoning is performed based on the interval confidence rule base. First, the input data is matched with the reference interval of each rule, and the corresponding rules are activated by calculating the matching degree of the input data in the reference interval. Then, the evidential reasoning (ER) method is used to integrate these activated rules and finally output the fault diagnosis results of the marine diesel engine. Step 5: Calculate the reliability of expert knowledge; Calculate the reliability of expert knowledge. This step evaluates the reliability of the initial model parameters constructed by expert knowledge to ensure that the credibility of expert knowledge is reasonably introduced into the model, thereby providing a reliable knowledge basis for subsequent reasoning. Step 6: Optimize the model and perform interpretability analysis; The model is optimized based on the reliability of the expert knowledge calculated in step 5. This step reasonably introduces the credibility of expert knowledge in the optimization process to ensure that the model not only relies on data during optimization, but also balances expert knowledge, thereby enhancing the interpretability of the model and further improving the accuracy and robustness of the diagnostic results. Step 7: Consider indicator uncertainty processing; Dealing with uncertainty in input data, by introducing a calculation method for attribute reliability, evaluating data reliability based on the fluctuation and noise of sensor data, and comprehensively considering data uncertainty in the reasoning process; Step 8: Robustness analysis of the model; Based on the results of uncertainty processing in step 7, further robustness analysis is performed to evaluate the stability of the model in the face of external disturbances such as noise and data fluctuations, ensuring that the model can still maintain accuracy and robustness in multi-source data and complex environments. Combined with the evaluation of attribute reliability, the robustness of fault diagnosis of the system under different working conditions is verified; Step 9: Obtaining the fault diagnosis result of the marine diesel engine; Obtain fault diagnosis results for marine diesel engines; By integrating the multi-source input data, expert knowledge and optimized reasoning of the model processed in the above steps, the fault diagnosis results of the marine diesel engine are finally output. This step is based on the reasoning process of the interval confidence rule base, combined with the analysis of uncertainty and robustness, to ensure that the diagnosis results have high accuracy and stability in complex working environments.

2. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: The types of vibration signal data samples acquired in step 1 include but are not limited to acceleration signals, velocity signals, displacement signals, and vibration frequency spectra.

3. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: The fault diagnosis index system constructed in step 2 also includes processing the measured data obtained in step 1. The processing flow first pre-processes the vibration signal to remove noise and interference signals to ensure the accuracy and stability of the data. Next, feature extraction is performed, and multi-dimensional analysis is performed by combining time domain features, frequency domain features, and time-frequency domain features: time domain features such as mean, variance, kurtosis, and skewness are used to measure the basic statistical characteristics of the vibration signal; frequency domain features analyze the main frequency, spectrum energy distribution, and power spectrum density of the vibration signal through frequency components to help identify the contribution of different frequency components to the fault; The time-frequency domain features capture the changes in signal time and frequency through methods such as wavelet transform and short-time Fourier transform. After feature extraction, these features are screened using the knowledge of domain experts, and key features closely related to the health status of marine diesel engines are selected to build an index system for fault diagnosis. This index system will include a series of numerical features that reflect different aspects of the bearing health status. These indicators will become the input for building an interval confidence rule base fault diagnosis model in subsequent steps.

4. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: In the step 3, an expert knowledge base and an index reference interval are constructed. First, based on the experience of experts in the field and historical fault data, key fault influencing factors in the operation of marine diesel engines are identified, and an expert knowledge base is constructed. The knowledge base includes information about various fault modes, operating states, and corresponding diagnostic characteristics. Subsequently, in the process of dividing the reference interval, each diagnostic indicator is divided into multiple reference intervals according to the fluctuation range of the actual measurement data and the statistical characteristics of the historical data; Each reference interval represents the range of variation of a specific indicator under different health states, and can effectively deal with the fluctuations and uncertainties of data during operation. Experts refine and adjust the upper and lower limits of these intervals based on their experience to ensure that each reference interval can accurately reflect the characteristic changes under different fault modes.

5. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: In the step 4, an inference model based on an interval confidence rule base is constructed. First, the initial parameters for constructing the interval confidence rule base can be obtained based on the step 3. Therefore, the Kth rule of the IF-THEN rule based on the interval confidence rule base is expressed as: where x i (i=1,...,M) represents the premise attribute fault feature in the input model, M is the number of premise attributes, [a i ,b i ] represents the reference interval set of the i-th indicator, where i=1,...,Mr k represents the reliability of the rule; w k represents the rule weight, L is the number of rules, β N,k Indicates D N The confidence level is the Nth failure mode, where N represents the number of failure modes. When a set of features (x1, x2, ..., x M )In the input model, each feature x i (1≤i≤M) falls within an interval, activating a rule. For feature x i (1≤i≤M), the rule matching degree calculation formula is: where a i represents the interval rule matching degree corresponding to the i-th attribute, Mid = (a i +b i ) / 2 indicates the position of the midpoint of the interval; After calculating the rule matching degree, the i-th feature x i The activation weight of the (1≤i≤M) activated rule is calculated as follows: w i =[ε+(1-ε)×]×λ i (3) Among them, the activation weight of the activated i-th interval rule is recorded as w i , initial rule weight λ i ; The minimum activation weight ε is given by expert knowledge and can be any constant in the range of 0 to 1; ER rules provide a transparent and intuitive framework for evidence combination. In addition, expert knowledge can be integrated, and the impact of different evidence on the entire reasoning process can be easily explained and evaluated. ER rules provide a reliable solution for decision-making and evaluation with multiple evidences. The rules in the model building process are actually used as evidence in the ER rules. L independent pieces of evidence are recorded as e i (i=1,...,L), the identification framework is denoted as Θ, which consists of N evaluation levels D n (n=1,...,N), which can be expressed as Θ={D1,...,D N }, then a piece of evidence can be represented as the following confidence distribution: Among them, β n,i is represented as the evaluation scheme in the evidence The following is evaluated as D n The confidence level, β Θ,i is represented as global ignorance, i.e., the confidence of the i-th premise attribute relative to the identification framework Θ; Assume that the weight of evidence is w i (i=1,...,L), and w i ∈[0,1]; the reliability of evidence is r i (i=1,...,L), and r i ∈[0,1]; then, the confidence distribution with mixed weighted evidence of reliability can be expressed as: Among them, the power set is denoted as β(Θ), and the i-th attribute is at level D n The probability mass of the mixture under It can be obtained by the following formula: Among them, the normalization coefficient is denoted as c rw,i =1 / (1+w i -r i ), which satisfies The joint support of any two pieces of evidence is β n,e(2) The calculation is as follows: Then, the joint support of L independent pieces of evidence β n,e(L) It can be calculated in the following general way: Where k = 3, 4, ..., L, β n,e(k) It is the first k attributes after fusion relative to level D n The confidence level, and m n,e(1) =m n,1 ,m β(Θ),e(1) =m β(Θ),1 , calculated by the above formula, the comprehensive evaluation results can be obtained as follows: e(L)={(D n ,b n,e(L) ),n=1,...,N,(Θ,β Θ,e(L) )} (13) Will be graded D n The utility of n ), then we can get the final output result y, that is, the expected utility value. The method for calculating the final expected utility value is as follows:

6. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: In step 5, the reliability of expert knowledge is calculated: Specifically, the reliability of expert knowledge is not only based on the initial parameters set by the expert, but also combined with objective analysis to quantify the credibility of expert knowledge. First, assuming that the observed data of the i-th attribute is x i (n)(n=1,2,...,N,i=1,2,...,M), then, the expert system constructed by expert knowledge constructs an interval confidence rule base to statistically evaluate the initial parameters and obtain the reliability evaluation value, the evaluation value r e The calculation of is based on comparing the predicted result with the actual state, using the mean squared error to quantify this difference: where k is the credibility of expert knowledge assessed through historical records and peer review, and r e =0 means that the expert knowledge is completely reliable.

7. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: Step 6 optimizes the model and performs interpretability analysis: The P-CMA-ES method is used to optimize the confidence, rule weight and rule reliability in the initial rule base to improve the accuracy of the state assessment model. Since the optimization process will have a negative impact on the interpretability of the method, interpretability constraints are added during the optimization process so that the optimization results do not violate the initial expert judgment and each optimized parameter maintains its original physical meaning. To build an optimization model, we must first clarify the function to be optimized. The function with interpretable optimization is expressed as: Among them, output model and output actual Respectively represent the predicted value and true value of the interval confidence rule base, T represents the number of data samples, (β, λ, r) initial represents the initial parameter values ​​set by expert knowledge, (β,λ,r) optimal represents the optimized parameters, (β,λ,r) low and (β,λ,r) up Represents the variation space of the parameters given by the expert; A reasonable and interpretable confidence distribution can be either monotonic or convex, but not concave:

8. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: The step 7 considers the indicator uncertainty processing. In engineering practice, the observed data may be disturbed, resulting in the observed data being unable to correctly reflect the system information, thus causing the data to be unreliable. The attribute reliability is calculated by setting a tolerance range to eliminate the interference of uncertain data. Let all inputs corresponding to the i-th attribute be x i,1 ,x i,2 ,...,x i,T ,i=1,2...,M means,x i,1 ,x i,2 ,...,x i,T The mean of The corresponding standard deviation is σ i ,i=1,2,...,M, so the tolerance range is Where λ is the adjustment factor of the tolerance range, which is usually determined based on expert knowledge. or When ij =1 means the observation is unreliable, otherwise u ij =0 means the observation is reliable. represents the number of unreliable observations, so the reliability of the i-th attribute is expressed as: Among them, the reliability of the i-th attribute is represented by h i (0≤h i ≤1) indicates that h i =1 means the attribute is completely reliable, h i =0 means completely unreliable.

9. The intelligent fault diagnosis method for marine diesel engines based on interval confidence rule base according to claim 1 is characterized in that: After the uncertainty is processed in step 7, the robustness of the model is further analyzed in step 8. In order to ensure the stability and reliability of the model under different input conditions, the Lipschitz condition is used to measure the disturbance amplitude and robustness of the model. Specifically, the response degree of the model to the input disturbance is calculated based on the Hatton distance to ensure that the model can remain robust in a complex environment. The Lipschitz condition of the interval confidence rule base is defined as follows: For M, The Interval Confidence Rule Model (IBRB) satisfies If there is a very small constant Make all This model is called Lipschitz stable, where d(·) is the distance metric, The disturbance The perturbation of x is represented by x′, and the Lipschitz condition at the Manhattan distance can be rewritten as: The modulo operation is represented by |·| The above definition gives the calculation of the Lipschitz constant for IBRB. Furthermore, for T, There is a functional relationship: If part of its function is As shown below: where x′ is the perturbation of x, c i is a small constant. When c i When ≤θ, we say that the function f is Lipschitz stable, and the corresponding Lipschitz constant is θ, where θ can be regarded as a perturbation constant. The smaller the constant, the more stable the model and the better the robustness. Taking the limit on both sides of equation (19) yields: At this time, the complex calculation of the Lipschitz constant is transformed into the calculation of the partial derivative of the high-dimensional function. By calculating the partial derivative of the high-dimensional function, the Lipschitz constant of the function relationship f can be calculated, and then the perturbation constant θ that makes the function f stable can be obtained, thereby realizing the robustness measurement of the function.

10. The marine diesel engine intelligent fault diagnosis method based on interval confidence rule base according to claim 1 is characterized in that: The ninth step obtains the fault diagnosis result of the marine diesel engine. After completing the reasoning, optimization and robustness analysis of the previous steps, the fault diagnosis result of the marine diesel engine is obtained. Through the comprehensive reasoning of input data processing, application of expert knowledge base, uncertainty processing and robustness analysis, the model finally outputs the fault diagnosis result of the engine. The result is not only based on the fusion analysis of multi-dimensional data, but also takes into account the uncertainty in the diagnosis process and the impact of complex environment on the data, ensuring the high accuracy and reliability of the fault diagnosis result.

Images