Fault analysis and prediction method applied to electric control system of electric forklift

Abstract

The invention discloses a fault analysis and prediction method for an electric control system of an electric forklift, and the method comprises the steps: enabling the electric control system to be divided into three modules: 1, an electric control system hierarchical division and fault coding module, enabling an electric control fault system to be divided into three layers A, B and C, carrying out the step-by-step coding according to a system, a subsystem, parts and a fault phenomenon, and carrying out the standardized coding according to an A + B + C rule; the fault positioning and handling module is used for tracing and positioning faults according to hierarchical codes, checking a fault reason probability value and checking and handling fault reasons in combination with a Bayesian tree structure and parallel message transmission; the trend analysis and fault prediction reasoning module is used for analyzing a timing sequence parameter trend of a B-level part, predicting a potential C-level fault and carrying out pretreatment maintenance; the method is used for improving the readability of fault phenomenon codes and the timeliness of fault analysis and positioning, and in addition, auxiliary analysis and troubleshooting are carried out on possible fault phenomena through the trend analysis and fault prediction reasoning module.

Description

Technical Field

[0001] This invention relates to the fields of fault decomposition and coding, fault analysis, and fault prediction in electric forklift control systems, and specifically to a fault analysis and prediction method applied to electric forklift control systems. Technical Background

[0002] Electric forklifts operate under complex conditions and require frequent use of the electronic control system. This necessitates that the electronic control system provide stable power output, quickly locate faults when they occur, or predict fault trends in advance to facilitate rapid handling or maintenance and prevent downtime from disrupting user work efficiency.

[0003] Currently, there are many suppliers of electronic control systems for new energy electric forklifts, and the fault code definitions of each supplier are inconsistent, making it difficult to trace the cause of the fault when the electronic control system malfunctions. When a fault is triggered, the existing fault symptom codes are only used to display the currently triggered fault symptom, and it is impossible to determine which specific subsystem or component is related to the fault based on the current symptom code, thus failing to narrow down the fault search scope and achieve rapid fault location and analysis. Summary of the Invention

[0004] The purpose of this invention is to provide a fault analysis and prediction method for electric forklift control systems, which improves the readability of fault codes and the timeliness of fault analysis and location. In addition, it uses trend analysis and fault prediction reasoning modules to assist in the analysis and troubleshooting of possible fault phenomena.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is: a fault analysis and prediction method applied to the electronic control system of an electric forklift, which divides the electronic control system into: an electronic control system hierarchical division and fault coding module, a fault location and handling module, and a trend analysis and fault prediction reasoning module;

[0006] The electronic control system hierarchy and fault coding module divides the electronic control system fault hierarchy into three levels: A, B and C. It classifies and codes the system, subsystem, components and fault phenomena level by level, and adopts the A+B+C method as a unified coding rule to standardize the coding of existing fault phenomena.

[0007] The fault location and handling module traces the source layer by layer according to the divided A, B, and C levels of code to locate the specific fault-related component fault phenomenon, narrows the search scope of fault-related components in level B, and uses Bayesian tree structure and parallel message passing to find multiple fault cause probability values ​​of the current fault-related components in level B, and investigates and handles the fault triggering factors one by one.

[0008] The trend analysis and fault prediction reasoning module performs trend analysis on the time-series change parameters of various fault-related components at level B. Based on the real-time change trend, it predicts potential level C fault phenomena and performs fault pre-processing and auxiliary maintenance measures.

[0009] Standardized fault symptom coding reduces the complexity of fault symptom localization and the scope of fault-related component operating parameter retrieval. It enables rapid reverse lookup and fault localization after fault symptom codes are generated. At the same time, it adds trend analysis and fault prediction to assist in positive fault trend prediction. By predicting possible fault symptom through parameter-based fault trend prediction, it allows for early handling of potential fault symptom.

[0010] Furthermore, in the electronic control system hierarchy and fault coding module, level A divides the electronic control system into drive control system and drive execution system; level B further refines level A, subdividing it into fault-related subsystems and fault-related components. Both levels B and C can expand fault-related subsystems and fault-related components, and assign codes according to unified rules; level C classifies and subdivides the fault phenomena related to the components divided in level B.

[0011] Furthermore, in the fault code encoding after the hierarchical division of the electronic control system shown, 3 digits are used to encode the relevant systems at level A; 5 digits are used to encode the relevant components at level B; and 4 digits are used to encode the relevant fault phenomena at level C. The complete fault phenomenon code is obtained by combining the codes of A, B, and C. The electronic control system monitors the uniformly compiled fault codes. If a certain fault code is triggered, the specific system and component are initially located based on the respective codes of levels A, B, and C.

[0012] Furthermore, the fault location and handling module specifically monitors the specific fault phenomena of C-level faults. If a fault phenomenon is triggered, it uses a Bayesian tree structure and a parallel message passing method to perform reverse reasoning. It calculates the posterior probability based on the conditional probability formula P(B-level fault-related components | C-level observed anomalies) and performs reverse reasoning to trace the potential influencing factors after the current fault phenomenon is triggered. This includes: Bayesian hierarchical causal structure construction, multi-threaded parallel computing, and reverse reasoning.

[0013] Furthermore, in the fault location and handling module, a Bayesian tree structure is constructed based on the fault phenomena corresponding to level C and the historical operating parameter data of the relevant components at level B:

[0014] (1) Leaf node design: Select the fault phenomenon after C-level triggering as the leaf node observation variable;

[0015] (2) Root cause node design: The maximum mutual information algorithm is used to screen the B-level fault-related component operation variables that are related to the fault phenomenon triggered at level C. The mutual information value is used to judge the correlation between the cause variable and the result variable. The larger the mutual information value, the stronger the correlation between the two. According to the (fault cause variable X and fault result variable Y) corresponding to levels B and C, the mutual information I(X; Y) calculation formula is constructed as shown in Equation 1):

[0016]

[0017] In the formula: P(x,y) is the joint probability distribution of X and Y, and P(x) and P(y) are the marginal probability distributions of X and Y.

[0018] The mutual information I(X;Y) is used to determine the operating parameters of the components related to the C-level fault phenomenon and the B-level fault. If I(X;Y) = 0, it means that X and Y are not related. The larger the value of I(X;Y), the stronger the correlation. A threshold α is designed to filter and record the operating variables of the B-level fault-related components with mutual information values ​​I(X;Y) > α, and these variables are stored in a data set X, as shown in Equation 2).

[0019] X = [x1, x2, x3, ..., x n ] 2)

[0020] Use set X as the root node of the Bayesian tree structure;

[0021] (3) State node design: The "intermediate state variables" connecting the root cause node and the leaf node are quantified and stored in the set Sa, as shown in Equation 3):

[0022] Sa=[Sa1, Sa2, Sa3,..., Sa n ] 3)

[0023] Where: Sa1, Sa2, Sa3, ..., Sa n The status of the operating parameters of the fault-related components corresponding to the current B level;

[0024] (4) Bayesian tree structure design: build a tree structure for the X and Sa sets and the C-level fault phenomenon node variables: treat the variables as nodes and the mutual information I(X;Y) as the "weight" of the edge; use the Prim algorithm to find the tree structure with the "maximum total weight" as the final Bayesian tree structure, and finally form a three-level tree structure of "root node → state node → leaf node", with only one path between any two nodes;

[0025] (5) Calculation of root cause analysis:

[0026] Parallel message passing is used for backward reasoning. The posterior probability is calculated based on the conditional probability formula P(operating parameters of B-level fault-related components | C-level observed anomalies). Backward reasoning is then used to trace the potential influencing factors after the current fault phenomenon is triggered. If a single C-level fault phenomenon is triggered, single-threaded diagnostic reasoning is performed; if multiple C-level fault phenomena are triggered simultaneously, multi-threaded parallel backward reasoning is performed. The result is the set of probability values ​​px for each variable included in the set of operating variables of B-level fault-related components, X, as shown in Equation 4).

[0027] px=[px1,px2,px3,…,px n ] 4)

[0028] The probabilities of the set px are sorted from largest to smallest to locate the most relevant B-level fault-related component operating parameters for fault handling.

[0029] Furthermore, the trend analysis and fault prediction inference module specifically: monitors the time-series operating parameters of components related to B-level faults in real time; removes noise and abnormal interference using outlier detection and LOESS local weighted regression methods; smooths operating parameters with large fluctuations; after data normalization, uses time-series decomposition methods to extract the trend direction, trend intensity, and abrupt change point trend characteristics of the operating parameters of individual B-level components; models and quantifies the operating trend of single parameters using traditional time-series models; predicts and warns of potential C-level fault phenomena based on the quantified B-level operating parameter trends; thus, before a fault occurs, it can proactively identify B-level components with unreasonable current operating trends, achieving fault prediction analysis.

[0030] Furthermore, in the trend analysis and fault prediction reasoning module, mutual information is used to determine the set of operating parameters X of the components related to the C-level fault phenomenon and the B-level fault, where X = [x1, x2, x3, ..., x...]. n The ordered execution parameters x1, x2, x3, ..., x in the set X. n After removing outliers and noise fluctuations, the time-series trend patterns of each parameter change are extracted. The specific steps are as follows:

[0031] (1) Data processing

[0032] Outlier handling: The 3σ method is used to handle multiple runtime variables x1, x2, x3, ..., xn in the variable set X. n Outlier detection and processing are performed separately to remove values ​​that exceed the mean ±3σ, thus avoiding false alarms from the sensor, such as sudden changes in trend caused by the sensor's instantaneous over-range value.

[0033] Missing value handling: For x1, x2, x3, ..., xn For missing data for each running variable, linear interpolation is used to fill in the missing data.

[0034] Data Fluctuation Smoothing: The LOESS local weighted regression method removes noise and abnormal interference, smooths out highly fluctuating operating parameters, locally fits the data, and preserves the operating trend. The LOESS local weighted regression model is shown in Equation 5).

[0035]

[0036] In the formula: x0 is the target point; x i For the runtime variables in the B-level runtime scalar set X; y j The fault phenomenon corresponding to level C is a result variable that includes fluctuations; d i For data point x i Distance from target point x0; d max The maximum distance;

[0037] The "influence of local data" is determined based on the weighting function, as shown in Formula 6):

[0038]

[0039] In the formula: w i (x0) is the weight function. If the weight w i =1 represents x i If it coincides with x0, and w i =0 represents x i Since they are outside the local domain and have less influence, the B-level runtime variable parameters with less influence can be filtered out.

[0040] For multiple runtime variables x1, x2, x3, ..., xn of variable set X n After data processing, the corresponding optimized variable set X is obtained. opt =[x 1opt x 2opt x 3opt ... x nopt ];

[0041] (2) Extraction of variable trend features and judgment of trend strength

[0042] For the set of variables X opt Chinese x 1opt x 2opt x 3opt ... x nopt Extract trend direction, trend strength, periodicity, and abrupt change features from relevant variable operating parameters;

[0043] Trend direction extraction: Construct a linear regression model, as shown in Equation 7):

[0044] y = kx + b 7)

[0045] In the formula: x is time, and y is the parameter value;

[0046] Linear trends are extracted using linear regression fitting: the slope k is used to determine the trend (k>0 indicates an upward trend, k<0 indicates a downward trend, and k≈0 indicates a stable trend).

[0047] Construct a multinomial regression model, as shown in Equation 8):

[0048] y = ax 2 +bx+c 8)

[0049] In the formula: x is time, and y is the parameter value;

[0050] Concavity / convexity is determined by multinomial regression fitting method;

[0051] The strength of the nonlinear trend is determined by the sum of squared residuals (SSE), as shown in Equation 9. The smaller the SSE, the better the fitted curve fits the data.

[0052]

[0053] Using the coefficient of determination R 2 To determine the strength of a linear trend, R 2 The closer R is to 1, the more pronounced the linear trend. 2 The calculation formula is shown in Equation 10.

[0054]

[0055] In the formula: The predicted value for the current parameter, y i The current running parameter value. The mean;

[0056] After extracting the trend features of the variables and judging the strength of the trend, X is obtained from the set of relevant variables at level B. opt =[x 1opt x 2opt x 3opt ... X nopt The trend of each variable in ];

[0057] By quantifying the trends of B-level operating parameters, potential C-level fault phenomena can be predicted and warned; thus, before a fault occurs, the B-level components corresponding to unreasonable current operating trends can be identified in advance, achieving fault prediction and analysis.

[0058] The beneficial effects of this invention are:

[0059] 1) This invention reduces the complexity of fault location by standardizing fault phenomenon coding, reduces the search range of operating parameters of fault-related components, and quickly realizes reverse lookup and fault location after the fault phenomenon code is generated, thereby improving the readability of fault phenomenon code and the timeliness of fault analysis and location.

[0060] 2) This invention adds trend analysis and fault prediction to assist in positive fault trend prediction. It predicts possible fault phenomena through parameter fault trend and takes measures in advance.

[0061] Attached image content

[0062] Figure 1 Overall framework for hierarchical division and fault coding of electronic control systems;

[0063] Figure 2 Overall framework for component fault location and handling;

[0064] Figure 3 Component operating parameter trend analysis and fault phenomenon prediction;

[0065] Figure 4 Fault symptom coding method after hierarchical division of electronic control system;

[0066] Figure 5 Bayesian fault tree structure for electronic control systems;

[0067] Figure 6 Overall process for fault analysis and prediction of electric forklift control system. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. However, it should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of the invention.

[0069] Unless otherwise defined, 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 pertains. The terminology used herein in the description of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0070] like Figure 1-6 As shown, a fault analysis and prediction method for electric forklift control systems divides the control system into: a control system hierarchy and fault coding module, a fault location and handling module, and a trend analysis and fault prediction reasoning module.

[0071] In the hierarchical division and fault coding module of the electronic control system, such as Figure 1As shown, the electronic control fault system is divided into three levels: Level A, Level B, and Level C. The levels are further refined and coded according to the system, subsystem, component, and fault phenomenon.

[0072] At level A, the electronic control system is divided into drive control system and drive execution system. At level B, the related systems involved in level A are further refined into fault-related subsystems and fault-related components. Both levels B and C can expand the fault-related subsystems and components, and a unified coding rule is used. At level C, the fault phenomena related to the components classified in level B are categorized and subdivided, such as logic type, status type, communication type, three-electric type, low-level type, and control port type. Within each category, the fault phenomena are further refined, such as input errors and over-temperature protection fault phenomena in the logic type, and stalled rotor and undervoltage fault phenomena in the status type. Each type of fault phenomenon can be expanded. Figure 4 The fault code encoding method shown in the diagram, after the hierarchical division of the electronic control system, uses 3 digits to encode the relevant systems at level A, such as "093" representing the drive control system and "092" representing the drive execution system; it uses 5 digits to encode the relevant components at level B, such as "02400" representing the drive motor-motor encoder; and it uses 4 digits to encode the relevant fault phenomena at level C, such as "9178" representing the status-related high speed fault phenomenon. Combining the codes from A, B, and C results in the complete fault code "092024009178," indicating that the motor encoder inside the drive execution system is reporting excessive speed. By monitoring the pre-compiled fault codes of the electronic control system, if a fault code is triggered, the specific system and component are initially located based on the respective codes of levels A, B, and C.

[0073] Example: If the currently detected fault code is "092024009178", first identify the "excessive speed" fault corresponding to the C-level fault code "9178", and then identify the drive motor-motor encoder related components corresponding to the B-level fault code "02400".

[0074] like Figure 2 As shown, the fault location and handling module traces the source layer by layer according to the divided A, B, and C levels of code to locate the specific fault-related component fault phenomenon, narrowing the search scope of fault-related components in level B, and using Bayesian tree structure and parallel message passing to find multiple fault cause probability values ​​of the current fault-related components in level B, and checking and handling the fault triggering factors one by one.

[0075] The fault location and handling module specifically monitors the specific fault phenomena of C-level faults. If a fault phenomenon is triggered, it uses a Bayesian tree structure and parallel message passing to perform reverse reasoning. It calculates the posterior probability based on the conditional probability formula P(B-level fault-related components | C-level observed anomalies) and performs reverse reasoning to trace the potential influencing factors after the current fault phenomenon is triggered. This includes: Bayesian hierarchical causal structure construction, multi-threaded parallel computing, and reverse reasoning.

[0076] In the fault location and handling module, a Bayesian tree structure is built based on the fault phenomena corresponding to level C and the historical operating parameter data of the relevant components at level B:

[0077] (1) Leaf node design: Select the fault phenomenon after C-level triggering as the leaf node observation variable;

[0078] (2) Root cause node design: The maximum mutual information algorithm is used to screen the B-level fault-related component operation variables that are related to the fault phenomenon triggered at level C. The mutual information value is used to judge the correlation between the cause variable and the result variable. The larger the mutual information value, the stronger the correlation between the two. According to the (fault cause variable X and fault result variable Y) corresponding to levels B and C, the mutual information I(X; Y) calculation formula is constructed as shown in Equation 1):

[0079]

[0080] In the formula: P(x,y) is the joint probability distribution of X and Y, and P(x) and P(y) are the marginal probability distributions of X and Y.

[0081] The mutual information I(X;Y) is used to determine the operating parameters of the components related to the C-level fault phenomenon and the B-level fault. If I(X;Y) = 0, it means that X and Y are not related. The larger the value of I(X;Y), the stronger the correlation. A threshold α is designed to filter and record the operating variables of the B-level fault-related components with mutual information values ​​I(X;Y) > α, and these variables are stored in a data set X, as shown in Equation 2).

[0082] X = [x1, x2, x3, ..., x n ] 2)

[0083] Use set X as the root node of the Bayesian tree structure;

[0084] (3) State node design: The "intermediate state variables" connecting the root cause node and the leaf node are quantified and stored in the set Sa, as shown in Equation 3):

[0085] Sa=[Sa1, Sa2, Sa3,..., Sa n ] 3)

[0086] Where: Sa1, Sa2, Sa3, ..., Sa n This refers to the operating parameter status of the fault-related components corresponding to the current B level, such as intermediate states like: current value too high, current less than threshold, severe overvoltage, etc.

[0087] (4) Bayesian Tree Structure Design: A tree structure is constructed for the sets X and Sa, and the fault phenomenon node variables at level C. Variables are treated as nodes, and mutual information I(X; Y) is considered as the "weights" of the edges. Using Prim's algorithm, the tree structure with the "maximum total weight" is found as the final Bayesian tree structure, ultimately forming a three-layer tree structure of "root node → state node → leaf node," where there is only one path between any two nodes. Figure 5 As shown;

[0088] (5) Calculation of root cause analysis:

[0089] Parallel message passing is used for backward reasoning. The posterior probability is calculated based on the conditional probability formula P(operating parameters of B-level fault-related components | C-level observed anomalies). Backward reasoning is then used to trace the potential influencing factors after the current fault phenomenon is triggered. If a single C-level fault phenomenon is triggered, single-threaded diagnostic reasoning is performed; if multiple C-level fault phenomena are triggered simultaneously, multi-threaded parallel backward reasoning is performed. The result is the set of probability values ​​px for each variable included in the set of operating variables of B-level fault-related components, X, as shown in Equation 4).

[0090] px=[px1,px2,px3,…,px n ] 4)

[0091] The probabilities of the set px are sorted from largest to smallest to locate the most relevant B-level fault-related component operating parameters for fault handling.

[0092] The trend analysis and fault prediction reasoning module performs trend analysis on the time-series change parameters of various fault-related components at level B. Based on the real-time change trend, it predicts potential level C fault phenomena and performs fault pre-processing and auxiliary maintenance measures.

[0093] In the trend analysis and fault prediction reasoning module, mutual information is used to determine the set of operating parameters X of the components related to the C-level fault phenomenon and the B-level fault, where X = [x1, x2, x3, ..., x...]. n The ordered execution parameters x1, x2, x3, ..., x in the set X. nAfter removing outliers and noise such as fluctuations, the temporal trend of each parameter change is extracted.

[0094] (1) Data processing

[0095] Outlier handling: The 3σ method is used to handle multiple runtime variables x1, x2, x3, ..., xn in the variable set X. n Outlier detection and processing are performed separately to remove values ​​that exceed the mean ±3σ, thus avoiding false alarms from the sensor, such as sudden changes in trend caused by the sensor's instantaneous over-range value.

[0096] Missing value handling: For x1, x2, x3, ..., x n For missing data for each running variable, linear interpolation is used to fill in the missing data.

[0097] Data Fluctuation Smoothing: The LOESS local weighted regression method removes noise and abnormal interference, smooths out highly fluctuating operating parameters, locally fits the data, and preserves the operating trend. The LOESS local weighted regression model is shown in Equation 5).

[0098]

[0099] In the formula: x0 is the target point; x i For the runtime variables in the B-level runtime scalar set X; y j The fault phenomenon corresponding to level C is a result variable that includes fluctuations; d i For data point x i Distance from target point x0; d max The maximum distance;

[0100] The "influence of local data" is determined based on the weighting function, as shown in Formula 6):

[0101]

[0102] In the formula: w i (x0) is the weight function. If the weight w i =1 represents x i If it coincides with x0, and w i =0 represents x i Since they are outside the local domain and have less influence, the B-level runtime variable parameters with less influence can be filtered out.

[0103] For multiple runtime variables x1, x2, x3, ..., xn of variable set X n After data processing, the corresponding optimized variable set X is obtained. opt =[x 1opt x 2opt x 3opt ... xnopt ];

[0104] (2) Extraction of variable trend features and judgment of trend strength

[0105] For the set of variables X opt Chinese x 1opt x 2opt x 3opt ... x nopt By analyzing relevant variables and operating parameters, features such as trend direction, trend strength, periodicity, and abrupt change points can be extracted.

[0106] Trend direction extraction: Construct a linear regression model, as shown in Equation 7):

[0107] y = kx + b 7)

[0108] In the formula: x is time, and y is the parameter value;

[0109] Linear trends are extracted using linear regression fitting: the slope k is used to determine the trend (k>0 indicates an upward trend, k<0 indicates a downward trend, and k≈0 indicates a stable trend).

[0110] Construct a multinomial regression model, as shown in Equation 8):

[0111] y = ax 2 +bx+c 8)

[0112] In the formula: x is time, and y is the parameter value;

[0113] Concavity / convexity is determined by multinomial regression fitting method;

[0114] The strength of the nonlinear trend is determined by the sum of squared residuals (SSE) as shown in Equation 9. The smaller the SSE, the better the fitted curve fits the data.

[0115]

[0116] Using the coefficient of determination R 2 To determine the strength of a linear trend, R 2 The closer R is to 1, the more pronounced the linear trend. 2 The calculation formula is shown in Equation 10);

[0117]

[0118] In the formula: The predicted value for the current parameter, y i The current running parameter value. The mean;

[0119] After extracting the trend features of the variables and judging the strength of the trend, X can be obtained from the set of relevant variables at level B. opt =[x1opt x 2opt x 3opt ... x nopt The trend of each variable in the [], such as variable x 1opt The corresponding trend set is x 1opt_tre =[x 1opt_tre1 x 1opt1opt_tre2 x 1opt1opt_tre3 ... x 1opt1opt_tren ].

[0120] By quantifying the trends of B-level operating parameters, potential C-level faults can be predicted and alerted. This allows for proactive troubleshooting of B-level components with unreasonable current operating trends before a fault occurs, thus achieving fault prediction and analysis.

[0121] 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 or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A fault analysis and prediction method applied to the electronic control system of an electric forklift, characterized in that: The electronic control system is divided into a hierarchical classification and fault coding module, a fault location and handling module, and a trend analysis and fault prediction reasoning module. The electronic control system hierarchy and fault coding module divides the electronic control system fault hierarchy into three levels: A, B and C. It classifies and codes the system, subsystem, components and fault phenomena level by level, and adopts the A+B+C method as a unified coding rule to standardize the coding of existing fault phenomena. The fault location and handling module traces the source layer by layer according to the divided A, B, and C levels of code to locate the specific fault-related component fault phenomenon, narrowing the search scope of fault-related components in level B, and using Bayesian tree structure and parallel message passing to find multiple fault cause probability values ​​of the current fault-related components in level B, and investigates and handles the fault triggering factors one by one. The trend analysis and fault prediction reasoning module performs trend analysis on the time-series change parameters of various fault-related components at level B. Based on the real-time change trend, it predicts potential level C fault phenomena and performs fault pre-processing and auxiliary maintenance measures.

2. The fault analysis and prediction method for an electric forklift control system according to claim 1, characterized in that: In the electronic control system hierarchy and fault coding module, level A divides the electronic control system into drive control system and drive execution system; level B further refines level A, subdividing it into fault-related subsystems and fault-related components. Both levels B and C can expand the fault-related subsystems and fault-related components, and assign codes according to unified rules; level C classifies and subdivides the fault phenomena related to the components divided in level B.

3. The method for fault analysis and prediction applied to the electronic control system of an electric forklift according to claim 2, characterized in that: In the fault code encoding after the hierarchical division of the electronic control system shown, 3 digits are used to encode the relevant systems at level A; 5 digits are used to encode the relevant components at level B; and 4 digits are used to encode the relevant fault phenomena at level C. The complete fault phenomenon code is obtained by combining the codes of A, B, and C. The electronic control system monitors the unified fault codes. If a certain fault code is triggered, the specific system and component are initially located based on the respective codes of levels A, B, and C.

4. The fault analysis and prediction method for an electric forklift control system according to claim 1, characterized in that: The specific fault location and handling module monitors the specific fault phenomena of C-level faults. If a fault phenomenon is triggered, it uses a Bayesian tree structure and parallel message passing method to perform reverse reasoning. It calculates the posterior probability based on the conditional probability formula P(B-level fault-related components | C-level observed anomalies) and performs reverse reasoning to trace the potential influencing factors after the current fault phenomenon is triggered. This includes: Bayesian hierarchical causal structure construction, multi-threaded parallel computing, and reverse reasoning.

5. A fault analysis and prediction method for an electric forklift control system according to claim 4, characterized in that: In the fault location and handling module, a Bayesian tree structure is built based on the fault phenomena corresponding to level C and the historical operating parameter data of the relevant components at level B: (1) Leaf node design: Select the fault phenomenon after C-level triggering as the leaf node observation variable; (2) Root Cause Node Design: The Maximum Mutual Information (MPI) algorithm is used to screen the B-level fault-related component operation variables that are related to the fault phenomenon triggered at level C. The mutual information value is used to determine the correlation between the cause variable and the result variable. The larger the mutual information value, the stronger the correlation between the two. Based on the (fault cause variable X and fault result variable Y) corresponding to levels B and C, the mutual information I(X) and Y(Y) calculation formulas are constructed as shown in Equation 1). In the formula: P(x,y) is the joint probability distribution of X and Y, and P(x) and P(y) are the marginal probability distributions of X and Y. The mutual information I(X;Y) is used to determine the operating parameters of the components related to the C-level fault phenomenon and the B-level fault. If I(X;Y) = 0, it means that X and Y are not related. The larger the value of I(X;Y), the stronger the correlation. A threshold α is designed to filter and record the operating variables of the B-level fault-related components with mutual information values ​​I(X;Y) > α, and these variables are stored in a data set X, as shown in Equation 2). X=[x1、x2、x3、…、x n ] 2) Use set X as the root node of the Bayesian tree structure; (3) State node design: The "intermediate state variables" connecting the root cause node and the leaf node are quantified and stored in the set Sa, as shown in Equation 3): Sa=[Sa1、Sa2、Sa3、…、Sa n 3) Where: Sa1, Sa2, Sa3, ..., Sa n The status of the operating parameters of the fault-related components corresponding to the current B level; (4) Bayesian tree structure design: build a tree structure for the X and Sa sets and the C-level fault phenomenon node variables: treat the variables as nodes and the mutual information I(X;Y) as the "weight" of the edge; use the Prim algorithm to find the tree structure with the "maximum total weight" as the final Bayesian tree structure, and finally form a three-level tree structure of "root node → state node → leaf node", with only one path between any two nodes; (5) Calculation of root cause analysis: The Parallel Message Passing method is used for reverse reasoning. The posterior probability is calculated based on the conditional probability formula P(operating parameters of B-level fault-related components | C-level observed anomalies). The potential influencing factors after the current fault phenomenon is triggered are traced back by reverse reasoning. If a C-level fault phenomenon is triggered, single-threaded diagnostic reasoning is performed. If multiple C-level fault phenomena are triggered at the same time, multi-threaded parallel reverse reasoning is performed. The probability value set px of each variable contained in the set of operational variables related to the faults of level B components is obtained, as shown in Equation 4): px=[px1、px2、px3、…、px n ] 4) The probabilities of the set px are sorted from largest to smallest to locate the most relevant B-level fault-related component operating parameters for fault handling.

6. A fault analysis and prediction method for an electric forklift control system according to claim 1, characterized in that: The trend analysis and fault prediction inference module specifically involves: real-time monitoring of the time-series operating parameters of components related to B-level faults; using outlier detection and LOESS local weighted regression methods to remove noise and abnormal interference; smoothing operating parameters with large fluctuations; and after data normalization, using time-series decomposition methods to extract the trend direction, trend intensity, and abrupt change point trend characteristics of the operating parameters of individual B-level components. The module then models and quantifies the operating trends of single parameters using traditional time-series models. Finally, it predicts and warns of potential C-level fault phenomena based on the quantified B-level operating parameter trends. This allows for early detection of B-level components that indicate unreasonable current operating trends before a fault occurs, enabling fault prediction and analysis.

7. A fault analysis and prediction method for an electric forklift control system according to claim 6, characterized in that: In the trend analysis and fault prediction reasoning module, mutual information is used to determine the set of operating parameters X of the components related to the C-level fault phenomenon and the B-level fault, where X = [x1, x2, x3, ..., x...]. n The ordered execution parameters x1, x2, x3, ..., x in the set X. n After removing outliers and noise fluctuations, the time-series trend patterns of each parameter change are extracted. The specific steps are as follows: (1) Data processing Outlier handling: The 3σ method is used to handle multiple runtime variables x1, x2, x3, ..., xn in the variable set X. n Outlier detection and processing are performed separately to remove values ​​that exceed the mean ±3σ, thus avoiding false alarms from the sensor, such as sudden changes in trend caused by the sensor's instantaneous over-range value. Missing value handling: For x1, x2, x3, ..., x n Missing data for each running variable is handled using... Linear interpolation is used for data filling. Data Fluctuation Smoothing: The LOESS local weighted regression method removes noise and abnormal interference, smooths out highly fluctuating operating parameters, locally fits the data, and preserves the operating trend. The LOESS local weighted regression model is shown in Equation 5). In the formula: x0 is the target point; x i For the runtime variables in the B-level runtime scalar set X; y j The fault phenomenon corresponding to level C is a result variable that includes fluctuations; d i d is the distance between data point x and target point x0; max The maximum distance; The "influence of local data" is determined based on the weighting function, as shown in Formula 6): In the formula: w i (x0) is the weight function. If the weight w i =1 represents x i If it coincides with x0, and w i =0 represents x i Since they are outside the local domain and have less influence, the B-level runtime variable parameters with less influence can be filtered out. For multiple runtime variables x1, x2, x3, ..., xn of variable set X n After data processing, the corresponding optimized variable set X is obtained. opt =[x 1opt x 2opt x 3opt ... x nopt ]; (2) Extraction of variable trend features and judgment of trend strength For the set of variables X opt Chinese x 1opt x 2opt x 3opt ... x nopt Extract trend direction, trend strength, periodicity, and abrupt change features from relevant variable operating parameters; Trend direction extraction: Construct a linear regression model, as shown in Equation 7): y = kx + b 7) In the formula: x is time, and y is the parameter value; Linear trends are extracted using linear regression fitting: the slope k is used to determine the trend (k>0 indicates an upward trend, k<0 indicates a downward trend, and k≈0 indicates a stable trend). Construct a multinomial regression model, as shown in Equation 8): y=ax 2 +bx+c 8) In the formula: x is time, and y is the parameter value; Concavity / convexity is determined by multinomial regression fitting method; The strength of the nonlinear trend is determined by the sum of squared residuals (SSE), as shown in Equation 9. The smaller the SSE, the better the fitted curve fits the data. Using the coefficient of determination R 2 To determine the strength of a linear trend, R 2 The closer R is to 1, the more pronounced the linear trend. 2 The calculation formula is shown in Equation 10. In the formula: The predicted value for the current parameter, y i The current running parameter value. The mean; After extracting the trend features of the variables and judging the strength of the trend, X is obtained from the set of relevant variables at level B. opt =[x 1opt x 2opt x 3opt ... x nopt The trend of each variable in ]; The potential fault phenomena at level C are predicted and warned by quantifying the trends of operating parameters at level B. This allows for early detection of B-level components that indicate unreasonable current operating trends before a fault occurs, enabling fault prediction and analysis.

Images