WO2024124658A1 - 基于案例学习与可诊断性分析的诊断算法量化推荐方法 - Google Patents

基于案例学习与可诊断性分析的诊断算法量化推荐方法 Download PDF

Info

Publication number
WO2024124658A1
WO2024124658A1 PCT/CN2023/072944 CN2023072944W WO2024124658A1 WO 2024124658 A1 WO2024124658 A1 WO 2024124658A1 CN 2023072944 W CN2023072944 W CN 2023072944W WO 2024124658 A1 WO2024124658 A1 WO 2024124658A1
Authority
WO
WIPO (PCT)
Prior art keywords
diagnosability
decision tree
fault diagnosis
tree model
recommendation
Prior art date
Application number
PCT/CN2023/072944
Other languages
English (en)
French (fr)
Inventor
陶来发
李商羽
王超
索明亮
丁宇
吕琛
于劲松
唐荻音
Original Assignee
北京航空航天大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京航空航天大学 filed Critical 北京航空航天大学
Publication of WO2024124658A1 publication Critical patent/WO2024124658A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound

Definitions

  • the present application relates to the technical field of fault diagnosis algorithm recommendation, and in particular to a diagnostic algorithm quantitative recommendation method based on case learning and diagnosability analysis.
  • Fault diagnosis algorithms are usually used to perform fault detection on objects to be detected.
  • There are many fault diagnosis algorithms and different fault diagnosis algorithms have different detection performance when performing fault detection on different objects to be detected.
  • different fault diagnosis algorithms may have different detection performance when performing fault detection on the same object to be detected; the same fault diagnosis algorithm may also have different detection performance when performing fault detection on different objects to be detected.
  • some fault diagnosis algorithms may be more suitable for detecting object A to be detected, and some fault diagnosis algorithms may be more suitable for detecting object B to be detected.
  • the purpose of this application is to provide a quantitative recommendation method for diagnostic algorithms based on case learning and diagnosability analysis, so as to reduce the influence of human subjective factors, improve the accuracy of fault diagnosis algorithm matching, reduce the workload of engineers, and improve matching efficiency.
  • this application proposes a technical paradigm for quantitative recommendation of fault diagnosis algorithms based on case learning, which achieves a substantial shift from qualitative recommendation to quantitative recommendation.
  • This technical paradigm has high practical engineering application value.
  • the recommendation results of the quantitative recommendation system of fault diagnosis algorithms based on case learning are explainable.
  • the decision tree recommender can better learn the applicable rules of various fault diagnosis algorithms and realize the effective recommendation of fault diagnosis algorithms.
  • the quantitative recommendation system of fault diagnosis algorithm based on case learning does not require expert experience and can avoid the subjective experience in traditional recommendation.
  • the human factor influence of judgment and qualitative analysis effectively absorbs the practical experience of existing diagnostic cases, avoids large-scale modeling and data testing, makes the recommendation process more generalized, the recommendation results more objective, and the recommendation decision-making more rapid and efficient.
  • an embodiment of the present application provides a diagnostic algorithm quantitative recommendation method based on case learning and diagnosability analysis, which may include:
  • diagnosability element information corresponding to each diagnosability element in the diagnosability element set of the object to be diagnosed Obtaining diagnosability element information corresponding to each diagnosability element in the diagnosability element set of the object to be diagnosed
  • the recommended decision tree model is the same as the field to which the object to be diagnosed belongs; in the recommended decision tree model, the object to be diagnosed is classified according to the diagnosability factor information to divide the object to be diagnosed into the category of the target fault diagnosis algorithm among multiple fault diagnosis algorithms.
  • the embodiment of the present application provides a first possible implementation of the first aspect, wherein the method may further include:
  • the training samples contain sample diagnosability element information corresponding to each of the diagnosability elements in the diagnosability element set of the sample diagnosis object, and the algorithm label of the sample fault diagnosis algorithm;
  • the sample fault diagnosis algorithm is the fault diagnosis algorithm used in the training sample when performing fault diagnosis for the sample diagnosis object; all the sample diagnosis objects belong to the same field;
  • the training samples corresponding to the fault diagnosis algorithm are divided according to a preset division ratio to obtain a first training set and a first test set corresponding to the fault diagnosis algorithm;
  • the second training set is a collection of the first training sets corresponding to all the fault diagnosis algorithms
  • the second test set is a collection of the first test sets corresponding to all the fault diagnosis algorithms
  • the initial decision tree model is determined as the recommended decision tree model.
  • the fault diagnosis algorithm used to train the recommended decision tree model may include a support vector machine (SVM), a K-nearest neighbor (KNN), a convolutional neural network (CNN), an autoencoder (AE), a hidden Markov model (HMM), and a shallow artificial neural network (ANN).
  • SVM support vector machine
  • KNN K-nearest neighbor
  • CNN convolutional neural network
  • AE autoencoder
  • HMM hidden Markov model
  • ANN shallow artificial neural network
  • the embodiment of the present application provides a second possible implementation of the first aspect, wherein the using the second training set to establish an initial decision tree model that meets the tree building cutoff condition may include:
  • the second diagnosability element with the smallest second Gini index is screened out from all the diagnosability elements included in the branch training set corresponding to the branch;
  • the split node is used as the current new root node, and the branch training set under each branch divided according to the cutting point of the root node and subsequent steps are repeatedly executed until the current decision tree model meets the tree building cutoff condition, and the current decision tree model that meets the tree building cutoff condition is used as the initial decision tree model.
  • an embodiment of the present application provides a third possible implementation of the first aspect, wherein the tree building cutoff condition may include any one or more of the following: the number of node layers in the current decision tree model reaches a preset number of layers; the second Gini index of the second diagnosability factor in the current decision tree model is less than a preset threshold; the number of training samples contained in the branch training set under the branch corresponding to the split point of the split node in the current decision tree model is less than a preset number.
  • the embodiment of the present application provides a fourth possible implementation of the first aspect, wherein the using the second test set to test the initial decision tree model and calculating the recommendation accuracy of the initial decision tree model may include:
  • a first number of correct recommendations and a second number of incorrect recommendations by the initial decision tree model are counted when the initial decision tree model is tested using the third test set, so as to determine, according to the first number and the second number, a first recommendation accuracy rate of the initial decision tree model when the initial decision tree model is tested using the third test set;
  • An average value of the first recommendation accuracy rates corresponding to each of the third test sets is calculated, and the average value is used as the recommendation accuracy rate of the initial decision tree model.
  • dividing the second test set to obtain multiple third test sets may include: evenly dividing the second test set so that the number of training samples contained in each of the third test sets is the same.
  • the first recommendation accuracy of the initial decision tree model may be a ratio of the first number to a third number, where the third number is the sum of the first number and the second number.
  • an embodiment of the present application provides a fifth possible implementation of the first aspect, wherein the specific engineering condition element set may include multiple of the following: industry engineering field, object type, object structure size, operating conditions, operating environment, operating conditions, and signal type, signal path number, data type, data volume, feature type, dimension of feature vector, and diagnosable fault type and diagnosable fault degree collected during the working process.
  • an embodiment of the present application provides a sixth possible implementation of the first aspect, wherein the fault diagnosis algorithm requirement element set may include any one or more of the following: fault diagnosis algorithm function, fault diagnosis algorithm complexity, fault diagnosis algorithm autonomy, and fault diagnosis algorithm performance.
  • the fault diagnosis algorithm performance may be any one or more of the following: the accuracy of the fault diagnosis algorithm diagnosis, the root mean square error of the fault diagnosis algorithm diagnosis, and the maturity of the fault diagnosis algorithm diagnosis.
  • the embodiment of the present application provides a method for quantitatively recommending diagnostic algorithms based on case study and diagnosability analysis.
  • all diagnosability factor information of the object to be diagnosed is input into a trained recommendation decision tree model.
  • the recommendation decision tree model classifies the object to be diagnosed according to the diagnosability factor information to classify the object to be diagnosed into the category of a target fault diagnosis algorithm among multiple fault diagnosis algorithms, thereby outputting a recommendation result corresponding to the target fault diagnosis algorithm.
  • this method recommends fault diagnosis algorithms to the object to be diagnosed through a trained recommendation decision tree model, which is conducive to reducing the influence of human subjective factors, improving the accuracy of fault diagnosis algorithm matching, reducing the workload of engineers, lowering the entry threshold for PHM engineers, and improving the efficiency of recommendation matching.
  • a second aspect of the present application further provides a computer-readable storage medium, in which instructions are stored, and when the instructions are executed by a processor, the processor can execute the method according to the above-mentioned embodiment of the present invention.
  • the third aspect of the present application also provides a computer software product, the computer software product is stored in a computer-readable storage medium, the computer software product includes instructions, the instructions are executed by a processor to enable The processor executes the method according to the above embodiment of the present invention.
  • FIG1 shows a flow chart of a diagnostic algorithm quantitative recommendation method based on case study and diagnosability analysis provided in an embodiment of the present application
  • FIG2 is a schematic diagram showing a diagnosability element set provided in an embodiment of the present application.
  • FIG3 shows a schematic diagram of a training sample provided in an embodiment of the present application
  • FIG4 shows a schematic diagram of a current decision tree model provided in an embodiment of the present application.
  • the embodiment of the present application provides a recommendation method for fault diagnosis algorithms based on a decision tree to reduce the influence of human subjective factors, improve the accuracy of fault diagnosis algorithm matching, reduce the workload of engineers, and improve matching efficiency.
  • the following is a description through an embodiment.
  • FIG1 shows a flowchart of a diagnostic algorithm quantitative recommendation method based on case learning and diagnosability analysis provided in an embodiment of the present application. As shown in FIG1 , the method may include the following steps S101-S102:
  • diagnosability factor information corresponding to each diagnosability factor in the diagnosability factor set of the object to be diagnosed.
  • Information diagnosability elements are related to fault diagnosis; the diagnosability element set includes the specific engineering condition element set and the fault diagnosis algorithm requirement element set.
  • the object to be diagnosed can specifically be a mechanical component to be diagnosed.
  • a gearbox For example, a gearbox.
  • FIG2 shows a schematic diagram of a diagnosability element set provided in an embodiment of the present application.
  • the specific engineering condition element set includes multiple of the following: industry engineering field, object type, object structure size, operating conditions, operating environment, and operating conditions, as well as the signal type, signal path number, data type, data volume, feature type, and dimension of the feature vector collected during the working process, as well as the diagnosable fault type and diagnosable fault degree.
  • the set of fault diagnosis algorithm requirement factors includes many of the following: fault diagnosis algorithm function, fault diagnosis algorithm complexity, fault diagnosis algorithm autonomy, and fault diagnosis algorithm performance.
  • the diagnosability factor information corresponding to the industry engineering field of the object to be diagnosed is: the industry engineering field to which the object to be diagnosed belongs, such as aircraft, wind turbine, mechanical vehicle, etc.
  • the diagnosability factor information corresponding to the object type of the object to be diagnosed is: which component the object to be diagnosed belongs to, such as a gear or a bearing.
  • the diagnosability factor information corresponding to the object structure size of the object to be diagnosed is: the size of the object to be diagnosed, the number of teeth, the transmission structure and other information.
  • the diagnosability factor information corresponding to the operating condition of the object to be diagnosed is: some quantitative operating condition information of the object to be diagnosed during the working process, such as rotation speed, workload, etc.
  • the diagnosability factor information corresponding to the usage environment of the object to be diagnosed is: the physical operating environment faced by the object to be diagnosed during actual operation, such as temperature, humidity, environmental vibration, environmental noise, environmental interference, etc.
  • the diagnosability factor information corresponding to the usage condition of the object to be diagnosed is: some background information not involved in the usage environment and usage conditions, such as constant speed conditions (single operating condition) or variable speed (multiple operating conditions) conditions, single fault type or multiple fault types and other qualitative descriptions.
  • the diagnosability factor information corresponding to the signal type of the object to be diagnosed is: the type of signal generated by the object to be diagnosed during the working process collected by various collection sensors and other collection devices.
  • the signal type includes vibration signal, current signal, torque signal, etc.
  • the diagnosability factor information corresponding to the number of signal paths of the object to be diagnosed is used to indicate: how many signals are collected by the collection device of the object to be diagnosed during the working process.
  • the collected signals are vibration signals, including vibration X-axis signals, vibration Y-axis signals, and vibration Z-axis signals, so a total of 3 signals are collected.
  • the diagnosability factor is the data type of the data collected during the work process
  • the data type of the object to be diagnosed The corresponding diagnosability element information is used to describe the structural characteristics of the data collected by the acquisition device during the working process of the object to be diagnosed, such as linear data, nonlinear data, etc.
  • the diagnosability factor information corresponding to the data amount of the object to be diagnosed is used to indicate the amount of data collected by the collection device during the working process of the object to be diagnosed.
  • the diagnosability factor information corresponding to the feature type of the object to be diagnosed is used to indicate: after collecting the data generated by the object to be diagnosed during the working process, the feature type proposed from the data, such as time domain features, frequency domain features or other features.
  • the diagnosability factor information corresponding to the dimension of the feature vector of the object to be diagnosed is used to indicate: after collecting the data generated by the object to be diagnosed during the working process and proposing a feature type from the data, the feature type is converted into a feature vector used to represent the feature type, and the dimension of the feature vector refers to the specific dimensionality of the feature vector.
  • the diagnosability factor information corresponding to the diagnosable fault type of the object to be diagnosed is: what specific fault types may occur in the object to be diagnosed.
  • the types of faults that may occur in the gearbox include: broken teeth, wear, crack faults of the gear, inner ring, outer ring, rolling element faults of the bearing, etc.
  • the diagnosability factor information corresponding to the degree of diagnosable fault of the object to be diagnosed is: what are the specific degrees of faults that may occur in the object to be diagnosed.
  • the object to be diagnosed is a gearbox, and the degrees of faults that may occur in the gearbox include: mild, moderate and severe.
  • the diagnosability factor is the function of the fault diagnosis algorithm
  • the diagnosability factor information corresponding to the function of the fault diagnosis algorithm of the object to be diagnosed is: the function of the target fault diagnosis algorithm that the decision tree model is expected to output.
  • the function of the algorithm that the decision tree model is expected to output is fault diagnosis.
  • the complexity of the fault diagnosis algorithm is the running time of the fault diagnosis algorithm, and/or the size of the storage space occupied when the fault diagnosis algorithm is running.
  • the running time of the fault diagnosis algorithm refers to the length of time required for the fault diagnosis algorithm to run
  • the size of the storage space occupied when the fault diagnosis algorithm is running refers to the size of the storage space temporarily occupied by the fault diagnosis algorithm during its operation.
  • the diagnosability factor is the complexity of the fault diagnosis algorithm
  • the diagnosability factor information corresponding to the complexity of the fault diagnosis algorithm of the object to be diagnosed is: within what range the complexity of the target fault diagnosis algorithm that the decision tree model is expected to output is (i.e. within what range the running time of the target fault diagnosis algorithm that the decision tree model is expected to output is expected to be, and/or within what range the size of the storage space temporarily occupied by the target fault diagnosis algorithm that the decision tree model is expected to output is expected to be).
  • the autonomy of the fault diagnosis algorithm is the number of hyperparameters in the fault diagnosis algorithm.
  • the number of hyperparameters is the number of parameters that can be manually adjusted during the use of the fault diagnosis algorithm to improve the algorithm's diagnostic ability.
  • the diagnosability factor is the autonomy of the fault diagnosis algorithm
  • the diagnosability factor information corresponding to the autonomy of the fault diagnosis algorithm of the object to be diagnosed is: The number of hyperparameters in the target fault diagnosis algorithm that you want the decision tree model to output.
  • the performance of the fault diagnosis algorithm is any one or more of the following: the accuracy of the fault diagnosis algorithm, the root mean square error of the fault diagnosis algorithm, and the maturity of the fault diagnosis algorithm.
  • the accuracy of the fault diagnosis algorithm represents the diagnostic ability of the fault diagnosis algorithm
  • the root mean square error of the fault diagnosis algorithm is used to measure the deviation between the observed value (the value diagnosed by the fault diagnosis algorithm) and the true value (actual value).
  • the maturity of the fault diagnosis algorithm may be evaluated by experts by referring to factors such as the generation time and usage of the fault diagnosis algorithm.
  • the diagnosability factor information corresponding to the performance of the fault diagnosis algorithm of the object to be diagnosed is: the performance of the target fault diagnosis algorithm that the decision tree model is expected to output, specifically, the accuracy of the target fault diagnosis algorithm that the decision tree model is expected to output, the root mean square error of the fault diagnosis algorithm is less than, the maturity of the fault diagnosis algorithm, etc.
  • S102 Input all diagnosability factor information into the trained recommendation decision tree model, output the recommendation result for the object to be diagnosed through the recommendation decision tree model, and perform fault diagnosis on the object to be diagnosed using the target fault diagnosis algorithm corresponding to the recommendation result.
  • the recommended decision tree model is the same as the field to which the object to be diagnosed belongs; in the recommended decision tree model, the object to be diagnosed is classified according to the diagnosability factor information to divide the object to be diagnosed into the category of the target fault diagnosis algorithm among multiple fault diagnosis algorithms.
  • the recommended decision tree model is the same as the engineering industry field (i.e., field) to which the object to be diagnosed belongs.
  • the engineering industry field to which the recommended decision tree model belongs is determined by the training samples of the recommended decision tree model, i.e., the engineering industry field to which the recommended decision tree model belongs is the same as the engineering industry field to which the training samples belong.
  • the recommendation decision tree model classifies the object to be diagnosed according to the diagnosability factor information to divide the object to be diagnosed into the category of the target fault diagnosis algorithm among multiple fault diagnosis algorithms, and then the recommendation decision tree model outputs the recommendation result corresponding to the target fault diagnosis algorithm according to the category of the divided target fault diagnosis algorithm.
  • the purpose of performing fault diagnosis on the object to be diagnosed is to determine whether the object to be diagnosed has a fault and the specific situation of the fault.
  • a recommendation method based on comprehensive conditions such as user needs, existing resource constraints, and algorithm capability constraints is proposed; a new paradigm for algorithm model recommendation is proposed; it is beneficial to reduce the influence of human subjective factors, improve the accuracy of fault diagnosis algorithm matching, reduce the workload of engineers, lower the entry threshold for PHM engineers, and improve the efficiency of recommendation matching.
  • step S101 before executing step S101 to obtain diagnosability element information corresponding to each diagnosability element in the diagnosability element set of the object to be diagnosed, the following may also be included:
  • S1001 Acquire multiple training samples corresponding to each fault diagnosis algorithm; the training samples contain sample diagnosis The sample diagnosability element information corresponding to each diagnosability element in the diagnosability element set of the object, and the algorithm label of the sample fault diagnosis algorithm; the sample fault diagnosis algorithm is the fault diagnosis algorithm used in the training sample when performing fault diagnosis for the sample diagnosis object; all sample diagnosis objects belong to the same field.
  • the fault diagnosis algorithm used to train the recommended decision tree model may include support vector machine (SVM), K nearest neighbor (KNN), convolutional neural network (CNN), autoencoder (AE), hidden Markov model (HMM), shallow artificial neural network (ANN).
  • SVM support vector machine
  • KNN K nearest neighbor
  • CNN convolutional neural network
  • AE autoencoder
  • HMM hidden Markov model
  • ANN shallow artificial neural network
  • FIG3 shows a schematic diagram of a training sample provided in an embodiment of the present application.
  • each fault diagnosis algorithm corresponds to a chart
  • each fault diagnosis algorithm corresponds to multiple training samples
  • each horizontal row in the chart represents a training sample.
  • the sample diagnosability factor information corresponding to the industry engineering field of the sample diagnostic object is: the industry engineering field to which the sample diagnostic object belongs.
  • the diagnosability factor is the object type
  • the diagnosability factor information corresponding to the object type of the sample diagnostic object is: which component the sample diagnostic object belongs to.
  • the diagnosability factor is the object structure size
  • the diagnosability factor information corresponding to the object structure size of the sample diagnostic object is: the size of the sample diagnostic object, the number of teeth, the transmission structure and other information.
  • the diagnosability factor is the operating condition
  • the diagnosability factor information corresponding to the operating condition of the sample diagnostic object is: some quantitative operating condition information of the object to be diagnosed during the working process.
  • the diagnosability factor is the use environment
  • the diagnosability factor information corresponding to the use environment of the sample diagnostic object is: some background information that is not involved in the use environment and the use condition of the object to be diagnosed.
  • the diagnosability factor information corresponding to the signal type of the sample diagnostic object is: the type of signal generated by the sample diagnostic object during operation.
  • the diagnosability factor is the number of signal paths
  • the diagnosability factor information corresponding to the number of signal paths of the sample diagnostic object is: how many signals are collected by the acquisition device during operation of the sample diagnostic object.
  • the diagnosability factor is a data type
  • the diagnosability factor information corresponding to the data type of the sample diagnostic object is: the type of data collected by the sample diagnostic object through the acquisition device during operation.
  • the diagnosability factor is data volume
  • the diagnosability factor information corresponding to the data volume of the sample diagnostic object is: the amount of data collected by the sample diagnostic object through the acquisition device during operation.
  • the diagnosability factor information corresponding to the feature type of the sample diagnostic object is: after collecting the data generated by the sample diagnostic object during the working process, the feature type is extracted from the data.
  • the diagnosability factor is the dimension of the feature vector
  • the diagnosability factor information corresponding to the dimension of the feature vector of the sample diagnostic object is used to indicate: after collecting the data generated by the sample diagnostic object during the working process and extracting the feature type from the data, the feature type is converted into a feature vector used to represent the feature type, and the dimension of the feature vector refers to the specific dimensionality of the feature vector.
  • the diagnosability factor information corresponding to the diagnosable fault type of the sample diagnostic object is: what specific fault types may occur in the sample diagnostic object.
  • the diagnosability factor information corresponding to the diagnosable fault degree of the sample diagnostic object is: what specific fault degrees may occur in the sample diagnostic object.
  • the diagnosability factor information corresponding to the fault diagnosis algorithm function of the sample diagnosis object is: the function of the sample fault diagnosis algorithm in the training sample.
  • the diagnosability factor information corresponding to the complexity of the fault diagnosis algorithm of the sample diagnosis object is: the complexity of the sample fault diagnosis algorithm in the training sample, that is, the running time of the sample fault diagnosis algorithm in the training sample, and/or the size of the temporarily occupied storage space.
  • the diagnosability factor is the autonomy of the fault diagnosis algorithm
  • the diagnosability factor information corresponding to the autonomy of the fault diagnosis algorithm of the sample diagnosis object is: the number of hyperparameters in the sample fault diagnosis algorithm in the training sample.
  • the diagnosability factor information corresponding to the performance of the fault diagnosis algorithm of the sample diagnosis object is: the accuracy, root mean square error, and maturity of the sample fault diagnosis algorithm in the training sample.
  • the instructions in the embodiment of step S101 refer to the instructions in the embodiment of step S101.
  • the sample diagnosis object is an object for which fault diagnosis has been performed at a historical moment, so in this embodiment, the fault diagnosis algorithm used when the fault diagnosis was performed for the sample diagnosis object at a historical moment is used as the sample fault diagnosis algorithm of the sample diagnosis object.
  • all sample diagnosis objects belong to the same industry field.
  • S1002 For each fault diagnosis algorithm, divide the training samples corresponding to the fault diagnosis algorithm according to a preset division ratio to obtain a first training set and a first test set corresponding to the fault diagnosis algorithm.
  • the first training set corresponding to the fault diagnosis algorithm contains 80 training samples
  • the first test set corresponding to the fault diagnosis algorithm contains 20 training samples.
  • S1003 Using the second training set to establish an initial decision tree model that meets the tree building cutoff condition; the second training set is the set of first training sets corresponding to all fault diagnosis algorithms.
  • the set of first training sets of each fault diagnosis algorithm is used as the second training set.
  • the first training set corresponding to each fault diagnosis algorithm includes 80 training samples
  • the second training set has a total of 480 (6 ⁇ 80) training samples.
  • the CART (Classification and Regression Tree) algorithm is used to establish an initial decision tree model that meets the tree building cutoff condition using the second training set.
  • the initial decision tree model in this embodiment is a CART initial decision tree model.
  • S1004 Use the second test set to test the initial decision tree model and calculate the recommendation accuracy of the initial decision tree model; the second test set is a collection of the first test sets corresponding to all fault diagnosis algorithms.
  • the set of the first test sets of all fault diagnosis algorithms is used as the second test set, the initial decision tree model is tested using the second test set, and the recommendation accuracy of the initial decision tree model is calculated.
  • the initial decision tree model is continuously trained until the recommendation accuracy of the initial decision tree model is greater than the preset accuracy.
  • the preset accuracy may be 70%-99%, and may be 75%.
  • step S1003 when executing step S1003 to use the second training set to establish an initial decision tree model that meets the tree building cutoff condition, the following steps may be specifically performed:
  • S10031 For each diagnosability factor information in the second training set, calculate the first Gini index of the diagnosability factor information.
  • the Gini value corresponding to the second training set is first calculated by the following formula:
  • Gini is the Gini value, which refers to the probability value of the diagnosability factors in two training samples randomly selected from the second training set, and the two diagnosability factors do not belong to the training samples corresponding to the same fault diagnosis algorithm.
  • the Gini value is intended to measure the "impurity" of the second training set. The smaller its value is, the higher the "purity" of the second training set is.
  • M is the total number of fault diagnosis algorithms, and M is 6 in this embodiment.
  • i represents the i-th fault diagnosis algorithm, and the value of i is an integer from 1 to 6.
  • pi is the ratio of the number of training samples corresponding to the i-th fault diagnosis algorithm to the total number of training samples in the second training set.
  • NS represents the total number of training samples in the second training set, and Ni represents the number of training samples corresponding to the i-th fault diagnosis algorithm.
  • the first Gini index of each diagnosability factor in the second training set is calculated by the following formula:
  • S represents the second training set
  • b represents the diagnosability factor
  • GiniIndex(S,b) is the first Gini index of the diagnosability factor b
  • M' represents the number of branches formed after the second training set is divided by the diagnosability factor b
  • K represents the Kth branch among the number of branches formed after the second training set is divided by the diagnosability factor b
  • NK represents the number of training samples contained in the Kth branch.
  • S10032 Based on all first Gini indexes, select the first diagnosability factor with the smallest first Gini index from all diagnosability factors.
  • S10033 Taking the first diagnosability element as a root node, and taking the first sample diagnosability element information corresponding to the first diagnosability element as a dividing point of the root node.
  • Figure 4 shows a schematic diagram of a current decision tree model provided by an embodiment of the present application.
  • the diagnosable fault degree of the sample diagnostic object if the first diagnosability factor with the smallest first Gini index among all diagnosability factors is the diagnosable fault degree of the sample diagnostic object, then the diagnosable fault degree of the sample diagnostic object is taken as the root node, and mild, moderate, and severe (mild, moderate, and severe are the diagnosability factor information of the diagnosable fault degree) are taken as the dividing points of the root node.
  • S10034 Obtain a branch training set under each branch divided according to the cutting point of the root node.
  • the branch training set under the branch corresponding to the split point is the training set composed of the training samples marked as 1 for mild in FIG. 3 .
  • the sample fault diagnosis algorithm when used to diagnose the sample diagnosis object, it is diagnosed multiple times at different historical moments. Therefore, when the diagnosis is performed at different moments, the diagnosis results under the same diagnosability factor may have multiple situations. For example, the diagnosable fault degree of sample diagnosis object a is mild at the first diagnosis, moderate at the second diagnosis, and severe at the third diagnosis. Therefore, the diagnosable fault degree of sample diagnosis object a included in the training sample includes mild, moderate, and severe.
  • S10035 For each branch training set, according to each diagnosability factor in the branch training set, calculate the second Gini index of the diagnosability factor.
  • step S10031 when calculating the second Gini index of each diagnosability factor in each branch training set, specific reference may be made to the calculation process in the embodiment of step S10031, which will not be repeated in this application.
  • S10037 Use the second diagnosability element as the splitting node of each branch, and use the second diagnosability element information corresponding to the second diagnosability element as the cutting point of the splitting node.
  • the signal type is used as the splitting node of the branch
  • the second diagnosability element information corresponding to the signal type i.e., vibration signal, torque signal, current signal
  • S10038 Determine whether the current decision tree model meets the tree building cutoff condition.
  • the tree building cutoff conditions include any one or more of the following: the number of node layers in the current decision tree model reaches a preset number of layers; the second Gini index of the second diagnosability factor in the current decision tree model is less than a preset threshold; the number of training samples contained in the branch training set under the branch corresponding to the split point of the split node in the current decision tree model is less than a preset number.
  • step S1004 when executing step S1004 to test the initial decision tree model using the second test set and calculating the recommendation accuracy of the initial decision tree model, the following steps may be specifically performed:
  • the second test set may be evenly divided so that the number of training samples contained in each third test set is the same.
  • the number of the third test sets is multiple.
  • the number of the third test sets may be 5.
  • S10043 Based on the sample recommendation results and algorithm labels corresponding to the same training sample, count a first number of correct recommendations and a second number of incorrect recommendations by the initial decision tree model when the initial decision tree model is tested using the third test set, so as to determine, based on the first number and the second number, the first recommendation accuracy of the initial decision tree model when the initial decision tree model is tested using the third test set.
  • the sample fault diagnosis type corresponding to the sample recommendation result recommended by the initial decision tree model for the sample diagnosis object when the sample fault diagnosis type corresponding to the sample recommendation result recommended by the initial decision tree model for the sample diagnosis object is the same as the algorithm label corresponding to the sample diagnosis object, it means that the initial decision tree model recommendation is correct.
  • the sample fault diagnosis type corresponding to the sample recommendation result recommended by the initial decision tree model for the sample diagnosis object is different from the algorithm label corresponding to the sample diagnosis object, it means that the initial decision tree model recommendation is wrong.
  • the first recommendation accuracy is the ratio of the first number to the third number, and the third number is the sum of the first number and the second number.
  • Each third test set corresponds to a first recommendation accuracy.
  • S10044 Calculate the average of the first recommendation accuracy rates corresponding to each third test set, and use the average value as the recommendation accuracy rate of the initial decision tree model.
  • each functional unit in the embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a processor-executable non-volatile computer-readable storage medium.
  • the technical solution of the present application can essentially be embodied in the form of a software product, or in other words, the part that contributes to the prior art or the part of the technical solution.
  • the computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application.
  • the aforementioned storage media include: USB flash drives, mobile hard drives, read-only memories (Read-Only Memory), etc. Memory (ROM), Random Access Memory (RAM), magnetic disk or CD-ROM, etc., which can store program code.
  • the present application provides a method for quantitatively recommending diagnostic algorithms based on case study and diagnosability analysis, including: obtaining diagnosability element information corresponding to each diagnosability element in the diagnosability element set of the object to be diagnosed; the diagnosability element is related to fault diagnosis; all diagnosability element information is input into a trained recommendation decision tree model, and the recommendation decision tree model outputs the recommendation result for the object to be diagnosed, so as to use the target fault diagnosis algorithm corresponding to the recommendation result to perform fault diagnosis on the object to be diagnosed; the recommendation decision tree model is the same as the field to which the object to be diagnosed belongs; in the recommendation decision tree model, the object to be diagnosed is classified according to the diagnosability element information, so as to classify the object to be diagnosed into the category of the target fault diagnosis algorithm.
  • This method is conducive to reducing the influence of human subjective factors, improving the accuracy of fault diagnosis algorithm matching, reducing the workload of engineers, and improving the efficiency of recommendation matching.
  • the diagnostic algorithm quantitative recommendation method based on case learning and diagnosability analysis of the present application is reproducible and can be used in a variety of industrial applications.
  • the diagnostic algorithm quantitative recommendation method based on case learning and diagnosability analysis of the present application can be used in the technical field of fault diagnosis algorithm recommendation.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Quality & Reliability (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

本申请提供了一种基于案例学习与可诊断性分析的诊断算法量化推荐方法，包括：获取待诊断对象的可诊断性要素集中各个可诊断性要素对应的可诊断性要素信息；可诊断性要素与故障诊断相关；将所有的可诊断性要素信息输入至训练好的推荐决策树模型中，通过推荐决策树模型输出针对待诊断对象的推荐结果，以使用推荐结果对应的目标故障诊断算法对待诊断对象进行故障诊断；推荐决策树模型与待诊断对象所属领域相同；在推荐决策树模型中，根据可诊断性要素信息对待诊断对象进行分类，以将待诊断对象划分至目标故障诊断算法的类别下。该方法有利于减少人为主观因素影响，提高故障诊断算法匹配的精确性，减少工程师的工作量，提高推荐匹配效率。

Description

基于案例学习与可诊断性分析的诊断算法量化推荐方法
相关申请的交叉引用
本申请要求于2022年12月14日提交中国国家知识产权局的申请号为202211598015.3、名称为“基于案例学习与可诊断性分析的诊断算法量化推荐方法”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及面向故障诊断算法推荐的技术领域，尤其是涉及一种基于案例学习与可诊断性分析的诊断算法量化推荐方法。
背景技术
故障诊断算法(故障预测与健康管理算法，PHM算法)通常用于对待检测对象进行故障检测，故障诊断算法有多种，不同的故障诊断算法对不同的待检测对象进行故障检测时，检测性能不同。具体地，不同的故障诊断算法对同一待检测对象进行故障检测时，检测性能可能不同；同一故障诊断算法对不同的待检测对象进行故障检测时，检测性能也可能不同。其中，有的故障诊断算法可能更适合检测待检测对象A，有的故障诊断算法更适合检测待检测对象B。
现有技术中，在为待检测对象选择故障诊断算法，并为其进行故障检测时，通常是工程师根据自己的工程经验进行选择，主观性较强。并且当同时存在多个待检测对象需要进行故障检测时，需要工程师逐一为各个待检测对象匹配故障诊断算法，导致工程师的工作量较大，匹配效率较低。
发明内容
有鉴于此，本申请的目的在于提供一种基于案例学习与可诊断性分析的诊断算法量化推荐方法，以减少人为主观因素影响，提高故障诊断算法匹配的精确性，以及减少工程师的工作量，提高匹配效率。
具体地，本申请提出了基于案例学习的故障诊断算法量化推荐技术范式，实现了从定性推荐到定量推荐的实质性转变，该技术范式具有高度实际工程应用价值。
基于案例学习的故障诊断算法量化推荐***推荐结果具有可解释性，其中决策树推荐器能够较好地学习到各类故障诊断算法的适用规则，可实现故障诊断算法的有效推荐。
基于案例学习的故障诊断算法量化推荐***无需专家经验，可以避免传统推荐中主观 判断和定性分析的人为因素影响，有效吸收了现有诊断案例的实践经验，避免了大规模的建模和数据测试，使得推荐过程更具泛化性，推荐结果更具有客观性，推荐决策更加迅速高效。
第一方面，本申请的实施例提供了一种基于案例学习与可诊断性分析的诊断算法量化推荐方法，可以包括：
获取待诊断对象的可诊断性要素集中各个可诊断性要素对应的可诊断性要素信息；
将所有的所述可诊断性要素信息输入至训练好的推荐决策树模型中，通过所述推荐决策树模型输出针对所述待诊断对象的推荐结果，以使用所述推荐结果对应的目标故障诊断算法对所述待诊断对象进行故障诊断；
其中，所述推荐决策树模型与所述待诊断对象所属领域相同；在所述推荐决策树模型中，根据所述可诊断性要素信息对所述待诊断对象进行分类，以将所述待诊断对象划分至多个故障诊断算法中的所述目标故障诊断算法的类别下。
结合第一方面，本申请实施例提供了第一方面的第一种可能的实施方式，其中，所述方法还可以包括：
获取各个所述故障诊断算法各自对应的多个训练样本；所述训练样本中包含有样本诊断对象的所述可诊断性要素集中各个所述可诊断性要素对应的样本可诊断性要素信息，以及样本故障诊断算法的算法标签；所述样本故障诊断算法是该训练样本中为该样本诊断对象进行故障诊断时使用的所述故障诊断算法；所有所述样本诊断对象所属领域相同；
针对每个所述故障诊断算法，按照预设划分比例，对该故障诊断算法对应的所述训练样本进行划分，得到该故障诊断算法对应的第一训练集和第一测试集；
使用第二训练集建立满足建树截止条件的初始决策树模型；所述第二训练集为所有所述故障诊断算法对应的所述第一训练集的集合；
使用第二测试集对所述初始决策树模型进行测试，计算所述初始决策树模型的推荐准确率；所述第二测试集为所有所述故障诊断算法对应的所述第一测试集的集合；
当所述推荐准确率大于预设准确率时，将所述初始决策树模型确定为所述推荐决策树模型。
在一些实施方式中，当推荐决策树模型所属的工程行业领域为齿轮箱时，则用于训练所述推荐决策树模型时所使用的故障诊断算法可以包括支持向量机(SVM)、K近邻(KNN)、卷积神经网络(CNN)、自编码器(AE)、隐马尔可夫模型(HMM)、浅层人工神经网络(ANN)。
结合第一方面的第一种可能的实施方式，本申请实施例提供了第一方面的第二种可能的实施方式，其中，所述使用第二训练集建立满足建树截止条件的初始决策树模型，可以包括：
针对所述第二训练集中的每个所述可诊断性要素，计算该可诊断性要素的第一基尼指数；
基于所有所述第一基尼指数，从所有所述可诊断性要素中筛选出第一基尼指数最小的第一可诊断性要素；
将第一可诊断性要素作为根节点，以及将所述第一可诊断性要素对应的第一样本可诊断性要素信息作为所述根节点的切分点；
获取按照根节点的切分点划分的每个分支下的分支训练集；
针对每个所述分支训练集，根据该分支训练集中的每个所述可诊断性要素，计算该可诊断性要素的第二基尼指数；
基于同一分支下的所有所述第二基尼指数，从该分支对应的所述分支训练集中包含的所有所述可诊断性要素中，筛选出第二基尼指数最小的第二可诊断性要素；
将所述第二可诊断性要素作为每个分支的***节点，并将所述第二可诊断性要素对应的第二样本可诊断性要素信息作为***节点的切分点；
判断当前决策树模型是否满足所述建树截止条件；
当所述当前决策树模型满足所述建树截止条件时，将所述当前决策树模型作为所述初始决策树模型；
当所述当前决策树模型不满足所述建树截止条件时，在所述当前决策树模型的基础上，将所述***节点作为当前新的根节点，重复执行获取按照根节点的切分点划分的每个分支下的分支训练集及后续步骤，直至当前决策树模型满足所述建树截止条件时停止，并将满足所述建树截止条件的当前决策树模型作为所述初始决策树模型。
结合第一方面的第二种可能的实施方式，本申请实施例提供了第一方面的第三种可能的实施方式，其中，所述建树截止条件可以包括以下中的任意一种或多种：所述当前决策树模型中节点层数达到预设层数；所述当前决策树模型中所述第二可诊断性要素的所述第二基尼指数小于预设阈值；所述当前决策树模型中所述***节点的切分点对应的分支下的分支训练集中包含的训练样本的数量小于预设数量。
结合第一方面的第一种可能的实施方式，本申请实施例提供了第一方面的第四种可能的实施方式，其中，所述使用第二测试集对所述初始决策树模型进行测试，计算所述初始决策树模型的推荐准确率，可以包括：
对所述第二测试集进行划分，得到多个第三测试集；
针对每个所述第三测试集中的各个所述可诊断性要素，将所述可诊断性要素对应的样本可诊断性要素信息输入到所述初始决策树中，通过所述初始决策树模型输出针对所述样本诊断对象的样本推荐结果；
根据同一训练样本对应的所述样本推荐结果和所述算法标签，统计出在使用所述第三测试集对所述初始决策树模型进行测试时，所述初始决策树模型推荐正确的第一数量和推荐错误的第二数量，以根据所述第一数量和所述第二数量，确定出在使用所述第三测试集对所述初始决策树模型进行测试时，所述初始决策树模型的第一推荐准确率；
计算各个所述第三测试集对应的所述第一推荐准确率的平均值，以将所述平均值作为所述初始决策树模型的所述推荐准确率。
在一些实施例中，对所述第二测试集进行划分，得到多个第三测试集可以包括：对所述第二测试集进行平均划分，使得每个所述第三测试集中包含的训练样本的数量相同。
在一些实施方式中，所述初始决策树模型的所述第一推荐准确率可以为所述第一数量与第三数量的比值，所述第三数量为所述第一数量与所述第二数量之和。
结合第一方面，本申请实施例提供了第一方面的第五种可能的实施方式，其中，所述特定工程条件要素集可以包括以下中的多种：所属行业工程领域、对象类型、对象结构尺寸、使用工况、使用环境、使用条件，以及在工作过程中采集到的信号类型、信号路数、数据类型、数据量、特征类型、特征向量的维数，以及可诊断故障类型、可诊断故障程度。
结合第一方面的第一种可能的实施方式，本申请实施例提供了第一方面的第六种可能的实施方式，其中，所述故障诊断算法需求要素集可以包括以下中的任意一种或多种：故障诊断算法功能、故障诊断算法复杂度、故障诊断算法自主性、故障诊断算法性能。
在一些实施例中，所述故障诊断算法性能可以为以下中的任意一种或多种：故障诊断算法诊断的准确率、故障诊断算法诊断的均方根误差、故障诊断算法诊断的成熟度。
本申请实施例提供的一种基于案例学习与可诊断性分析的诊断算法量化推荐方法，该实施例中，将待诊断对象的所有可诊断性要素信息输入至训练好的推荐决策树模型中，推荐决策树模型根据可诊断性要素信息对待诊断对象进行分类，以将待诊断对象划分至多个故障诊断算法中的目标故障诊断算法的类别下，从而输出与目标故障诊断算法对应的推荐结果。该方法与现有技术中工程师根据自身的工程经验为待诊断对象匹配出故障诊断算法的方式相比，本实施例通过训练好的推荐决策树模型对待诊断对象推荐故障诊断算法，有利于减少人为主观因素影响，提高故障诊断算法匹配的精确性，以及减少工程师的工作量，降低PHM工程师准入门槛，提高推荐匹配效率。
本申请的第二方面还提供了一种计算机可读取存储介质，所述计算机可读取存储介质中存储有指令，所述指令在由处理器执行时能够使所述处理器执行根据本发明的上述实施例所述的方法。
本申请的第三方面还还提供了一种计算机软件产品，所述计算机软件产品存储在计算机可读取存储介质中，所述计算机软件产品包含指令，所述指令在由处理器执行时能够使 所述处理器执行根据本发明的上述实施例所述的方法。
为使本申请的上述目的、特征和优点能更明显易懂，下文特举较佳实施例，并配合所附附图，作详细说明如下。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，应当理解，以下附图仅示出了本申请的某些实施例，因此不应被看作是对范围的限定，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他相关的附图。
图1示出了本申请实施例所提供的一种基于案例学习与可诊断性分析的诊断算法量化推荐方法的流程图；
图2示出了本申请实施例所提供的一种可诊断性要素集的示意图；
图3示出了本申请实施例所提供的一种训练样本的示意图；
图4示出了本申请实施例所提供的一种当前决策树模型的示意图。
具体实施方式
为使本申请实施例的目的、技术方案和优点更加清楚，下面将结合本申请实施例中附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。通常在此处附图中描述和示出的本申请实施例的组件可以以各种不同的配置来布置和设计。因此，以下对在附图中提供的本申请的实施例的详细描述并非旨在限制要求保护的本申请的范围，而是仅仅表示本申请的选定实施例。基于本申请的实施例，本领域技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本申请保护的范围。
考虑到现有技术中，工程师在根据自己的工程经验为待检测对象匹配故障诊断算法时，会造成主观性较强、工作量较大、匹配效率较低的问题，基于此，本申请实施例提供了一种基于决策树的故障诊断算法的推荐方法，以减少人为主观因素影响，提高故障诊断算法匹配的精确性，以及减少工程师的工作量，提高匹配效率。下面通过实施例进行描述。
为便于对本实施例进行理解，首先对本申请实施例所公开的一种基于决策树的故障诊断算法的推荐方法进行详细介绍。图1示出了本申请实施例所提供的一种基于案例学习与可诊断性分析的诊断算法量化推荐方法的流程图，如图1所示，该方法可以包括以下步骤S101-S102：
S101：获取待诊断对象的可诊断性要素集中各个可诊断性要素对应的可诊断性要素信 息；可诊断性要素与故障诊断相关；可诊断性要素集包括特定工程条件要素集和故障诊断算法需求要素集。
该实施例中，待诊断对象具体可以为待诊断的机械部件。例如齿轮箱。图2示出了本申请实施例所提供的一种可诊断性要素集的示意图，如图2所示，特定工程条件要素集包括以下中的多种：所属行业工程领域、对象类型、对象结构尺寸、使用工况、使用环境、使用条件，以及在工作过程中采集到的信号类型、信号路数、数据类型、数据量、特征类型、特征向量的维数，以及可诊断故障类型、可诊断故障程度。
故障诊断算法需求要素集包括以下中的多种：故障诊断算法功能、故障诊断算法复杂度、故障诊断算法自主性、故障诊断算法性能。
其中，当可诊断性要素为所属行业工程领域时，待诊断对象的所属行业工程领域对应的可诊断性要素信息为：待诊断对象所属的行业工程领域，例如飞机类、风力机类、机械车辆类等。
当可诊断性要素为对象类型时，待诊断对象的对象类型对应的可诊断性要素信息为：待诊断对象属于哪种部件，例如齿轮或者轴承等。
当可诊断性要素为对象结构尺寸时，待诊断对象的对象结构尺寸对应的可诊断性要素信息为：待诊断对象的尺寸大小、齿数多少、传动结构等信息。
当可诊断性要素为使用工况时，待诊断对象的使用工况对应的可诊断性要素信息为：待诊断对象在工作过程中的一些定量工况信息，例如转速、工作负载等。
当可诊断性要素为使用环境时，待诊断对象的使用环境对应的可诊断性要素信息为：待诊断对象在实际运行过程中所面对的物理运行环境，比如温度、湿度、环境振动、环境噪声、环境干扰等。
当可诊断性要素为使用条件时，待诊断对象的使用条件对应的可诊断性要素信息为：在使用环境、使用工况中没有涉及到的一些背景信息，比如恒速条件(单工况)或者变速度(多工况)条件、单故障类型或者多故障类型等定性描述。
当可诊断性要素为在工作过程中采集到的信号的信号类型时，待诊断对象的信号类型对应的可诊断性要素信息为：通过各类采集传感器等采集设备采集到的，待诊断对象在工作过程中产生的信号的类型。其中，信号类型包括振动信号、电流信号、扭矩信号等。
当可诊断性要素为在工作过程中采集到的信号的信号路数时，待诊断对象的信号路数对应的可诊断性要素信息用于表示：待诊断对象在工作过程中采集设备一共采集到了几路信号，例如采集的是振动信号，具体包括振动X轴信号，振动Y轴信号，振动Z轴信号，那么共采集有3路信号。
当可诊断性要素为在工作过程中采集到的数据的数据类型时，待诊断对象的数据类型 对应的可诊断性要素信息用于描述待诊断对象在工作过程中采集设备采集到的数据的结构特点，例如线性数据、非线性数据等。
当可诊断性要素为在工作过程中采集到的数据的数据量时，待诊断对象的数据量对应的可诊断性要素信息用于表示待诊断对象在工作过程中通过采集设备采集到的数据的数量。
当可诊断性要素为特征类型时，待诊断对象的特征类型对应的可诊断性要素信息用于表示：在采集到待诊断对象在工作过程中产生的数据后，从该数据中提出的特征类型，例如时域特征、频域特征或者其他特征等。
当可诊断性要素为特征向量的维数时，待诊断对象的特征向量的维数对应的可诊断性要素信息用于表示：在采集到待诊断对象在工作过程中产生的数据，且从该数据中提出的特征类型后，将该特征类型转换成用于表示该特征类型的特征向量，特征向量的维数指得是该特征向量具体是几维向量。
当可诊断性要素为可诊断故障类型时，待诊断对象的可诊断故障类型对应的可诊断性要素信息为：该待诊断对象可能发生的故障类型具体为哪些。例如，待诊断对象为齿轮箱，该齿轮箱可能发生的故障的类型包括：齿轮的断齿、磨损、裂纹故障，轴承的内圈、外圈、滚动体故障等。
当可诊断性要素为可诊断故障程度时，待诊断对象的可诊断故障程度对应的可诊断性要素信息为：该待诊断对象可能发生的故障程度具体为哪些。例如，待诊断对象为齿轮箱，该齿轮箱可能发生的故障的程度包括：轻度、中度和重度。当可诊断性要素为故障诊断算法功能时，待诊断对象的故障诊断算法功能对应的可诊断性要素信息为：希望决策树模型输出的目标故障诊断算法的功能。该实施例中，由于该待检测对象需要的算法为故障诊断算法，因此，希望决策树模型输出的算法的功能为故障诊断。
该实施例中，故障诊断算法复杂度为故障诊断算法的运行时间，和/或，故障诊断算法运行时占用存储空间大小。故障诊断算法运行时间是指故障诊断算法在运行过程中所需要的时长，故障诊断算法运行时占用存储空间大小是指故障诊断算法在运行过程中临时占用存储空间大小。当可诊断性要素为故障诊断算法复杂度时，待诊断对象的故障诊断算法复杂度对应的可诊断性要素信息为：希望决策树模型输出的目标故障诊断算法的复杂度在什么范围内(即希望决策树模型输出的目标故障诊断算法的运行时间在什么范围内，和/或，希望决策树模型输出的目标故障诊断算法在运行过程中临时占用存储空间大小在什么范围内)。
故障诊断算法自主性为故障诊断算法中超参数的数量。超参数的数量为该故障诊断算法在使用过程中人为可以调整的参数数量，从而来提高算法的诊断能力。当可诊断性要素为故障诊断算法自主性时，待诊断对象的故障诊断算法自主性对应的可诊断性要素信息为： 希望决策树模型输出的目标故障诊断算法中超参数的数量的多少。
故障诊断算法性能为以下中的任意一种或多种：故障诊断算法诊断的准确率、故障诊断算法诊断的均方根误差、故障诊断算法诊断的成熟度。其中，故障诊断算法诊断的准确率表征故障诊断算法的诊断能力，故障诊断算法诊断的均方根误差用于衡量观测值(故障诊断算法诊断出的值)与真实值(实际值)之间的偏差。故障诊断算法诊断的成熟度可能是专家通过参考该故障诊断算法的生成时间、使用情况等因素进行评价的。当可诊断性要素为故障诊断算法性能时，待诊断对象的故障诊断算法性能对应的可诊断性要素信息为：希望决策树模型输出的目标故障诊断算法的性能情况，具体地，希望决策树模型输出的目标故障诊断算法的准确率达到多少，故障诊断算法诊断的均方根误差小于多少，故障诊断算法诊断的成熟度为多少等。
S102：将所有的可诊断性要素信息输入至训练好的推荐决策树模型中，通过推荐决策树模型输出针对待诊断对象的推荐结果，以使用推荐结果对应的目标故障诊断算法对待诊断对象进行故障诊断。
其中，推荐决策树模型与待诊断对象所属领域相同；在推荐决策树模型中，根据可诊断性要素信息对待诊断对象进行分类，以将待诊断对象划分至多个故障诊断算法中的目标故障诊断算法的类别下。
该实施例中，推荐决策树模型与待诊断对象所属的工程行业领域(即领域)相同，推荐决策树模型所属的工程行业领域是由该推荐决策树模型的训练样本决定的，即推荐决策树模型所属的工程行业领域与训练样本所属的工程行业领域相同。
将可诊断性要素信息输入至训练好的推荐决策树模型中之后，推荐决策树模型根据可诊断性要素信息对待诊断对象进行分类，以将待诊断对象划分至多个故障诊断算法中的目标故障诊断算法的类别下，然后推荐决策树模型根据划分出的目标故障诊断算法的类别，输出与目标故障诊断算法对应的推荐结果。
该实施例中，对待诊断对象进行故障诊断的目的是为了确定该待诊断对象是否存在故障，以及故障的具体情况。
该实施例中，针对差异化故障诊断算法，提出了面向用户需求、现有资源约束、算法能力约束等综合条件下的推荐方法；提出了算法模型推荐的新范式；有利于减少人为主观因素影响，提高故障诊断算法匹配的精确性，以及减少工程师的工作量，降低PHM工程师准入门槛，提高推荐匹配效率。
在一种可能的实施方式中，在执行步骤S101获取待诊断对象的可诊断性要素集中各个可诊断性要素对应的可诊断性要素信息之前，还可以包括：
S1001：获取各个故障诊断算法各自对应的多个训练样本；训练样本中包含有样本诊断 对象的可诊断性要素集中各个可诊断性要素对应的样本可诊断性要素信息，以及样本故障诊断算法的算法标签；样本故障诊断算法是该训练样本中为该样本诊断对象进行故障诊断时使用的故障诊断算法；所有样本诊断对象所属领域相同。
该实施例中，以齿轮箱为例，当推荐决策树模型所属的工程行业领域为齿轮箱时，则用于训练该推荐决策树模型时所使用的故障诊断算法可以包括支持向量机(SVM)、K近邻(KNN)、卷积神经网络(CNN)、自编码器(AE)、隐马尔可夫模型(HMM)、浅层人工神经网络(ANN)。以上6类故障诊断算法覆盖了数据驱动的机器学习与深度学习经典分类算法，可解决不同情况下的齿轮箱故障诊断问题。
图3示出了本申请实施例所提供的一种训练样本的示意图，如图3所示，每个故障诊断算法对应一张图表，每个故障诊断算法对应多个训练样本，图表中每一横排表示一个训练样本。
当可诊断性要素为所属行业工程领域时，样本诊断对象的所属行业工程领域对应的样本可诊断性要素信息为：样本诊断对象所属的行业工程领域。当可诊断性要素为对象类型时，样本诊断对象的对象类型对应的可诊断性要素信息为：样本诊断对象属于哪种部件。当可诊断性要素为对象结构尺寸时，样本诊断对象的对象结构尺寸对应的可诊断性要素信息为：样本诊断对象的尺寸大小、齿数多少、传动结构等信息。当可诊断性要素为使用工况时，样本诊断对象的使用工况对应的可诊断性要素信息为：待诊断对象在工作过程中的一些定量工况信息。当可诊断性要素为使用环境时，样本诊断对象的使用环境对应的可诊断性要素信息为：待诊断对象在使用环境、使用工况中没有涉及到的一些背景信息。
当可诊断性要素为信号类型时，样本诊断对象的信号类型对应的可诊断性要素信息为：样本诊断对象在工作过程中产生的信号的类型。当可诊断性要素为信号路数时，样本诊断对象的信号路数对应的可诊断性要素信息为：样本诊断对象在工作过程中采集设备一共采集到了几路信号。当可诊断性要素为数据类型时，样本诊断对象的数据类型对应的可诊断性要素信息为：样本诊断对象在工作过程中通过采集设备采集到的数据的类型。当可诊断性要素为数据量时，样本诊断对象的数据量对应的可诊断性要素信息为：样本诊断对象在工作过程中通过采集设备采集到的数据的数量。
当可诊断性要素为特征类型时，样本诊断对象的特征类型对应的可诊断性要素信息为：在采集到样本诊断对象在工作过程中产生的数据后，从该数据中提取出的特征类型。当可诊断性要素为特征向量的维数时，样本诊断对象的特征向量的维数对应的可诊断性要素信息用于表示：在采集到样本诊断对象在工作过程中产生的数据，且从该数据中提取出的特征类型后，将该特征类型转换成用于表示该特征类型的特征向量，特征向量的维数指得是该特征向量具体是几维向量。
当可诊断性要素为可诊断故障类型时，样本诊断对象的可诊断故障类型对应的可诊断性要素信息为：该样本诊断对象可能发生的故障类型具体为哪些。当可诊断性要素为可诊断故障程度时，样本诊断对象的可诊断故障程度对应的可诊断性要素信息为：该样本诊断对象可能发生的故障程度具体为哪些。
当可诊断性要素为故障诊断算法功能时，样本诊断对象的故障诊断算法功能对应的可诊断性要素信息为：该训练样本中样本故障诊断算法的功能。当可诊断性要素为故障诊断算法复杂度时，样本诊断对象的故障诊断算法复杂度对应的可诊断性要素信息为：该训练样本中样本故障诊断算法的复杂度，即训练样本中样本故障诊断算法的运行时间，和/或，临时占用存储空间大小。当可诊断性要素为故障诊断算法自主性时，样本诊断对象的故障诊断算法自主性对应的可诊断性要素信息为：该训练样本中样本故障诊断算法中的超参数的数量。当可诊断性要素为故障诊断算法性能时，样本诊断对象的故障诊断算法性能对应的可诊断性要素信息为：该训练样本中样本故障诊断算法的准确率、均方根误差、成熟度。具体说明参考步骤S101实施例中的说明。
该实施例中，样本诊断对象为在历史时刻进行了故障诊断的对象，因此该实施例中，将历史时刻为该样本诊断对象进行故障诊断时使用的故障诊断算法作为该样本诊断对象的样本故障诊断算法。本实施例中，所有样本诊断对象所属的行业领域相同。
S1002：针对每个故障诊断算法，按照预设划分比例，对该故障诊断算法对应的训练样本进行划分，得到该故障诊断算法对应的第一训练集和第一测试集。
示例性的，当故障诊断算法“支持向量机”对应的训练样本的数量为100个，预设划分比例为8:2时，则对该故障诊断算法对应的训练样本进行划分后，得到该故障诊断算法对应的第一训练集中包含有80个训练样本，该故障诊断算法对应的第一测试集中包含有20个训练样本。
S1003：使用第二训练集建立满足建树截止条件的初始决策树模型；第二训练集为所有故障诊断算法对应的第一训练集的集合。
该实施例中，针对所有的故障诊断算法(即包括支持向量机、K近邻、卷积神经网络、自编码器、隐马尔可夫模型、浅层人工神经网络)，将各个故障诊断算法的第一训练集的集合作为第二训练集。例如，每个故障诊断算法对应的第一训练集中均包括80个训练样本，则第二训练集中共有480(6×80)个训练样本。
通过CART(Classification and Regression Tree)算法，使用第二训练集建立满足建树截止条件的初始决策树模型。本实施例中的初始决策树模型为CART初始决策树模型。
S1004：使用第二测试集对初始决策树模型进行测试，计算初始决策树模型的推荐准确率；第二测试集为所有故障诊断算法对应的第一测试集的集合。
将所有的故障诊断算法的第一测试集的集合作为第二测试集，使用第二测试集对初始决策树模型进行测试，计算初始决策树模型的推荐准确率。
S1005：当推荐准确率大于预设准确率时，将初始决策树模型确定为推荐决策树模型。
当推荐准确率不大于预设准确率时，继续对初始决策树模型进行训练，直至初始决策树模型的推荐准确率大于预设准确率。该实施例中预设准确率可以为70％-99％。具体的可以为75％。
在一种可能的实施例中，在执行步骤S1003使用第二训练集建立满足建树截止条件的初始决策树模型时，具体可以按照以下步骤执行：
S10031：针对第二训练集中的每个可诊断性要素信息，计算该可诊断性要素信息的第一基尼指数。
该实施例中，首先通过以下公式计算第二训练集对应的基尼值：
其中，Gini(S)为基尼值，基尼值是指从第二训练集中随机选出两个训练样本中的可诊断性要素，这两个可诊断性要素不属于同一个故障诊断算法对应的训练样本的概率值。基尼值旨在度量第二训练集的“不纯度”，其数值越小说明第二训练集的“纯净度”越高。M为故障诊断算法的总数量，该实施例中M为6。i表示第i个故障诊断算法，i的取值为1至6中的整数。pi为第i个故障诊断算法对应的训练样本的数量占第二训练集中训练样本的总数量的比值。NS表示第二训练集中训练样本的总数量，Ni表示第i个故障诊断算法对应的训练样本的数量。
通过以下公式计算第二训练集中的每个可诊断性要素的第一基尼指数：
其中，S表示第二训练集，b表示可诊断性要素，GiniIndex(S,b)为可诊断性要素b的第一基尼指数，M’表示通过可诊断性要素b对第二训练集划分后形成的分支数，K表示通过可诊断性要素b对第二训练集划分后形成的分支数中第K个分支。NK表示第K个分支下包含的训练样本的数量。
S10032：基于所有第一基尼指数，从所有可诊断性要素中筛选出第一基尼指数最小的第一可诊断性要素。
S10033：将第一可诊断性要素作为根节点，以及将第一可诊断性要素对应的第一样本可诊断性要素信息作为根节点的切分点。
示例性的，图4示出了本申请实施例所提供的一种当前决策树模型的示意图，如图4所示，若所有可诊断性要素中第一基尼指数最小的第一可诊断性要素为样本诊断对象的可诊断故障程度，那么将样本诊断对象的可诊断故障程度作为根节点，将轻度、中度、重度(轻度、中度、重度为可诊断故障程度的可诊断性要素信息)作为该根节点的切分点。
S10034：获取按照根节点的切分点划分的每个分支下的分支训练集。
承接步骤S10033中的实施例，切分点为轻度时，该切分点对应的分支下的分支训练集为图3中轻度标记为1的训练样本构成的训练集。
值得注意的是，在历史时刻，使用样本故障诊断算法对样本诊断对象进行故障诊断时，是在不同的历史时刻诊断了多次，因此，在不同时刻进行诊断时，同一个可诊断性要素下的诊断结果可能有多种情况。例如，第一次诊断时样本诊断对象a的可诊断故障程度为轻度，第二次诊断可能为中度，第三次诊断可能为重度，因此训练样本中包含的样本诊断对象a的可诊断故障程度即包括了轻度，又包括了中度和重度。
S10035：针对每个分支训练集，根据该分支训练集中的每个可诊断性要素，计算该可诊断性要素的第二基尼指数。
该实施例中，在计算每个分支训练集中的每个可诊断性要素的第二基尼指数时，具体可以参考步骤S10031的实施例中的计算过程，本申请对此不在重复叙述。
S10036：基于同一分支下的所有第二基尼指数，从该分支对应的分支训练集中包含的所有可诊断性要素中，筛选出第二基尼指数最小的第二可诊断性要素。
S10037：将第二可诊断性要素作为每个分支的***节点，并将第二可诊断性要素对应的第二可诊断性要素信息作为***节点的切分点。
如图3所示，承接步骤S10034中的实施例，切分点轻度对应的分支下的所有可诊断性要素中，假设第二基尼指数最小的第二可诊断性要素为信号类型，则将信号类型作为该分支的***节点，将信号类型对应的第二可诊断性要素信息(即振动信号、扭矩信号、电流信号)作为***节点信号类型的切分点。
S10038：判断当前决策树模型是否满足建树截止条件。
其中，建树截止条件包括以下中的任意一种或多种：当前决策树模型中节点层数达到预设层数；当前决策树模型中第二可诊断性要素的第二基尼指数小于预设阈值；当前决策树模型中***节点的切分点对应的分支下的分支训练集中包含的训练样本的数量小于预设数量。
S10039：当当前决策树模型满足建树截止条件时，将当前决策树模型作为初始决策树模型。
S100310：当当前决策树模型不满足建树截止条件时，在当前决策树模型的基础上，将 ***节点作为当前新的根节点，重复执行获取按照根节点的切分点划分的每个分支下的分支训练集及后续步骤，直至当前决策树模型满足建树截止条件时停止，并将满足建树截止条件的当前决策树模型作为初始决策树模型。
在一种可能的实施例中，在执行步骤S1004使用第二测试集对初始决策树模型进行测试，计算初始决策树模型的推荐准确率时，具体可以按照以下步骤执行：
S10041：对第二测试集进行划分，得到多个第三测试集。
该实施例中，可以对第二测试集进行平均划分，使得每个第三测试集中包含的训练样本的数量相同。第三测试集的数量为多个。示例性的，第三测试集的数量可以为5个。
S10042：针对每个第三测试集中的各个可诊断性要素，将该可诊断性要素对应的样本可诊断性要素信息输入到初始决策树中，通过初始决策树模型输出针对样本诊断对象的样本推荐结果。
S10043：根据同一训练样本对应的样本推荐结果和算法标签，统计出在使用第三测试集对初始决策树模型进行测试时，初始决策树模型推荐正确的第一数量和推荐错误的第二数量，以根据第一数量和第二数量，确定出在使用第三测试集对初始决策树模型进行测试时，初始决策树模型的第一推荐准确率。
该实施例中，当初始决策树模型为样本诊断对象推荐出的样本推荐结果对应的样本故障诊断类型，与该样本诊断对象对应的算法标签相同，则表示初始决策树模型推荐正确。当初始决策树模型为样本诊断对象推荐出的样本推荐结果对应的样本故障诊断类型，与该样本诊断对象对应的算法标签不同，则表示初始决策树模型推荐错误。
第一推荐准确率为第一数量与第三数量的比值，第三数量为第一数量与第二数量之和。每个第三测试集对应一个第一推荐准确率。
S10044：计算各个第三测试集对应的第一推荐准确率的平均值，以将平均值作为初始决策树模型的推荐准确率。
在本申请所提供的实施例中，应该理解到，所揭露的方法，可以通过其它的方式实现。
另外，在本申请实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个处理器可执行的非易失的计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only  Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
最后应说明的是：以上所述实施例，仅为本申请的具体实施方式，用以说明本申请的技术方案，而非对其限制，本申请的保护范围并不局限于此，尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化，或者对其中部分技术特征进行等同替换；而这些修改、变化或者替换，并不使相应技术方案的本质脱离本申请实施例技术方案的精神和范围，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。
工业实用性
本申请提供了一种基于案例学习与可诊断性分析的诊断算法量化推荐方法，包括：获取待诊断对象的可诊断性要素集中各个可诊断性要素对应的可诊断性要素信息；可诊断性要素与故障诊断相关；将所有的可诊断性要素信息输入至训练好的推荐决策树模型中，通过推荐决策树模型输出针对待诊断对象的推荐结果，以使用推荐结果对应的目标故障诊断算法对待诊断对象进行故障诊断；推荐决策树模型与待诊断对象所属领域相同；在推荐决策树模型中，根据可诊断性要素信息对待诊断对象进行分类，以将待诊断对象划分至目标故障诊断算法的类别下。该方法有利于减少人为主观因素影响，提高故障诊断算法匹配的精确性，减少工程师的工作量，提高推荐匹配效率。
此外，可以理解的是，本申请的基于案例学习与可诊断性分析的诊断算法量化推荐方法是可以重现的，并且可以用在多种工业应用中。例如，本申请的基于案例学习与可诊断性分析的诊断算法量化推荐方法可以用于面向故障诊断算法推荐的技术领域。

Claims (13)

  1. 一种基于案例学习与可诊断性分析的诊断算法量化推荐方法，其特征在于，包括：
    获取待诊断对象的可诊断性要素集中各个可诊断性要素对应的可诊断性要素信息；所述可诊断性要素与故障诊断相关；所述可诊断性要素集包括特定工程条件要素集和故障诊断算法需求要素集；
    将所有的所述可诊断性要素信息输入至训练好的推荐决策树模型中，通过所述推荐决策树模型输出针对所述待诊断对象的推荐结果，以使用所述推荐结果对应的目标故障诊断算法对所述待诊断对象进行故障诊断；
    其中，所述推荐决策树模型与所述待诊断对象所属领域相同；在所述推荐决策树模型中，根据所述可诊断性要素信息对所述待诊断对象进行分类，以将所述待诊断对象划分至多个故障诊断算法中的所述目标故障诊断算法的类别下。
  2. 根据权利要求1所述方法，其特征在于，还包括：
    获取各个所述故障诊断算法各自对应的多个训练样本；所述训练样本中包含有样本诊断对象的所述可诊断性要素集中各个所述可诊断性要素对应的样本可诊断性要素信息，以及样本故障诊断算法的算法标签；所述样本故障诊断算法是该训练样本中为该样本诊断对象进行故障诊断时使用的所述故障诊断算法；所有所述样本诊断对象所属领域相同；
    针对每个所述故障诊断算法，按照预设划分比例，对该故障诊断算法对应的所述训练样本进行划分，得到该故障诊断算法对应的第一训练集和第一测试集；
    使用第二训练集建立满足建树截止条件的初始决策树模型；所述第二训练集为所有所述故障诊断算法对应的所述第一训练集的集合；
    使用第二测试集对所述初始决策树模型进行测试，计算所述初始决策树模型的推荐准确率；所述第二测试集为所有所述故障诊断算法对应的所述第一测试集的集合；
    当所述推荐准确率大于预设准确率时，将所述初始决策树模型确定为所述推荐决策树模型。
  3. 根据权利要求2所述方法，其特征在于，当推荐决策树模型所属的工程行业领域为齿轮箱时，则用于训练所述推荐决策树模型时所使用的故障诊断算法包括支持向量机(SVM)、K近邻(KNN)、卷积神经网络(CNN)、自编码器(AE)、隐马尔可夫模型(HMM)、浅层人工神经网络(ANN)。
  4. 根据权利要求2或3所述方法，其特征在于，所述使用第二训练集建立满足建树截止条件的初始决策树模型，包括：
    针对所述第二训练集中的每个所述可诊断性要素，计算该可诊断性要素的第一基尼指 数；
    基于所有所述第一基尼指数，从所有所述可诊断性要素中筛选出第一基尼指数最小的第一可诊断性要素；
    将第一可诊断性要素作为根节点，以及将所述第一可诊断性要素对应的第一样本可诊断性要素信息作为所述根节点的切分点；
    获取按照根节点的切分点划分的每个分支下的分支训练集；
    针对每个所述分支训练集，根据该分支训练集中的每个所述可诊断性要素，计算该可诊断性要素的第二基尼指数；
    基于同一分支下的所有所述第二基尼指数，从该分支对应的所述分支训练集中包含的所有所述可诊断性要素中，筛选出第二基尼指数最小的第二可诊断性要素；
    将所述第二可诊断性要素作为每个分支的***节点，并将所述第二可诊断性要素对应的第二样本可诊断性要素信息作为***节点的切分点；
    判断当前决策树模型是否满足所述建树截止条件；
    当所述当前决策树模型满足所述建树截止条件时，将所述当前决策树模型作为所述初始决策树模型；
    当所述当前决策树模型不满足所述建树截止条件时，在所述当前决策树模型的基础上，将所述***节点作为当前新的根节点，重复执行获取按照根节点的切分点划分的每个分支下的分支训练集及后续步骤，直至当前决策树模型满足所述建树截止条件时停止，并将满足所述建树截止条件的当前决策树模型作为所述初始决策树模型。
  5. 根据权利要求4所述方法，其特征在于，所述建树截止条件包括以下中的任意一种或多种：所述当前决策树模型中节点层数达到预设层数；所述当前决策树模型中所述第二可诊断性要素的所述第二基尼指数小于预设阈值；所述当前决策树模型中所述***节点的切分点对应的分支下的分支训练集中包含的训练样本的数量小于预设数量。
  6. 根据权利要求2所述方法，其特征在于，所述使用第二测试集对所述初始决策树模型进行测试，计算所述初始决策树模型的推荐准确率，包括：
    对所述第二测试集进行划分，得到多个第三测试集；
    针对每个所述第三测试集中的各个所述可诊断性要素，将所述可诊断性要素各自对应的样本可诊断性要素信息输入到所述初始决策树中，通过所述初始决策树模型输出针对所述样本诊断对象的样本推荐结果；
    根据同一训练样本对应的所述样本推荐结果和所述算法标签，统计出在使用所述第三测试集对所述初始决策树模型进行测试时，所述初始决策树模型推荐正确的第一数量和推荐错误的第二数量，以根据所述第一数量和所述第二数量，确定出在使用所述第三测试集 对所述初始决策树模型进行测试时，所述初始决策树模型的第一推荐准确率；
    计算各个所述第三测试集对应的所述第一推荐准确率的平均值，以将所述平均值作为所述初始决策树模型的所述推荐准确率。
  7. 根据权利要求6所述方法，其特征在于，对所述第二测试集进行划分，得到多个第三测试集包括：对所述第二测试集进行平均划分，使得每个所述第三测试集中包含的训练样本的数量相同。
  8. 根据权利要求6或7所述方法，其特征在于，所述初始决策树模型的所述第一推荐准确率为所述第一数量与第三数量的比值，所述第三数量为所述第一数量与所述第二数量之和。
  9. 根据权利要求1至8中任一项所述方法，其特征在于，所述特定工程条件要素集包括以下中的多种：所属行业工程领域、对象类型、对象结构尺寸、使用工况、使用环境、使用条件，以及在工作过程中采集到的信号类型、信号路数、数据类型、数据量、特征类型、特征向量的维数，以及可诊断故障类型、可诊断故障程度。
  10. 根据权利要求1至9中任一项所述方法，其特征在于，所述故障诊断算法需求要素集包括以下中的多种：故障诊断算法功能、故障诊断算法复杂度、故障诊断算法自主性、故障诊断算法性能。
  11. 根据权利要求10所述方法，其特征在于，所述故障诊断算法性能为以下中的任意一种或多种：故障诊断算法诊断的准确率、故障诊断算法诊断的均方根误差、故障诊断算法诊断的成熟度。
  12. 一种计算机可读取存储介质，所述计算机可读取存储介质中存储有指令，所述指令在由处理器执行时能够使所述处理器执行根据权利要求1至11中任一项所述的方法。
  13. 一种计算机软件产品，所述计算机软件产品存储在计算机可读取存储介质中，所述计算机软件产品包含指令，所述指令在由处理器执行时能够使所述处理器执行根据权利要求1至11中任一项所述的方法。
PCT/CN2023/072944 2022-12-14 2023-01-18 基于案例学习与可诊断性分析的诊断算法量化推荐方法 WO2024124658A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211598015.3 2022-12-14
CN202211598015.3A CN115658371B (zh) 2022-12-14 2022-12-14 基于案例学习与可诊断性分析的诊断算法量化推荐方法

Publications (1)

Publication Number Publication Date
WO2024124658A1 true WO2024124658A1 (zh) 2024-06-20

Family

ID=85023706

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/072944 WO2024124658A1 (zh) 2022-12-14 2023-01-18 基于案例学习与可诊断性分析的诊断算法量化推荐方法

Country Status (2)

Country Link
CN (1) CN115658371B (zh)
WO (1) WO2024124658A1 (zh)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108009643A (zh) * 2017-12-15 2018-05-08 清华大学 一种机器学习算法自动选择方法和***
US20200242483A1 (en) * 2019-01-30 2020-07-30 Intuit Inc. Method and system of dynamic model selection for time series forecasting

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11775850B2 (en) * 2016-01-27 2023-10-03 Microsoft Technology Licensing, Llc Artificial intelligence engine having various algorithms to build different concepts contained within a same AI model
US11544494B2 (en) * 2017-09-28 2023-01-03 Oracle International Corporation Algorithm-specific neural network architectures for automatic machine learning model selection
CN110162014A (zh) * 2019-05-29 2019-08-23 上海理工大学 一种集成多种智能算法的制冷***故障诊断方法

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108009643A (zh) * 2017-12-15 2018-05-08 清华大学 一种机器学习算法自动选择方法和***
US20200242483A1 (en) * 2019-01-30 2020-07-30 Intuit Inc. Method and system of dynamic model selection for time series forecasting

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
ALI, S. ; SMITH, K.A.: "On learning algorithm selection for classification", APPLIED SOFT COMPUTING, ELSEVIER, AMSTERDAM, NL, vol. 6, no. 2, 1 January 2006 (2006-01-01), AMSTERDAM, NL , pages 119 - 138, XP024959590, ISSN: 1568-4946, DOI: 10.1016/j.asoc.2004.12.002 *
KOTLAR MILOS; PUNT MARIJA; RADIVOJEVIC ZAHARIJE; CVETANOVIC MILOS; MILUTINOVIC VELJKO: "Novel Meta-Features for Automated Machine Learning Model Selection in Anomaly Detection", IEEE ACCESS, IEEE, USA, vol. 9, 21 June 2021 (2021-06-21), USA , pages 89675 - 89687, XP011862900, DOI: 10.1109/ACCESS.2021.3090936 *

Also Published As

Publication number Publication date
CN115658371B (zh) 2023-03-10
CN115658371A (zh) 2023-01-31

Similar Documents

Publication Publication Date Title
CN107122594B (zh) 一种新能源车辆电池的健康预测方法和***
Li et al. A novel deep autoencoder and hyperparametric adaptive learning for imbalance intelligent fault diagnosis of rotating machinery
Zheng et al. A rolling bearing fault diagnosis method based on multi-scale fuzzy entropy and variable predictive model-based class discrimination
Wang et al. Random convolutional neural network structure: An intelligent health monitoring scheme for diesel engines
CN112418277A (zh) 旋转机械零部件剩余寿命预测方法、***、介质、设备
Wang et al. An intelligent fault diagnosis approach for planetary gearboxes based on deep belief networks and uniformed features
CN113642508B (zh) 基于参数自适应vmd与优化svm的轴承故障诊断方法
Arul et al. Data anomaly detection for structural health monitoring of bridges using shapelet transform
Jin et al. Fault diagnosis of rotating machines based on EEMD-MPE and GA-BP
Zhang et al. Graph neural network-based bearing fault diagnosis using Granger causality test
Özkan et al. Effect of data preprocessing on ensemble learning for classification in disease diagnosis
Sari et al. Performance Analysis of Resampling and Ensemble Learning Methods on Diabetes Detection as Imbalanced Dataset
Saha et al. Enhancing bearing fault diagnosis using transfer learning and random forest classification: A comparative study on variable working conditions
CN115931359B (zh) 一种涡轮泵轴承故障诊断方法及装置
Yadav et al. Variable dropout one-dimensional CNN for vibration-based shaft unbalance detection in industrial machinery
WO2024124658A1 (zh) 基于案例学习与可诊断性分析的诊断算法量化推荐方法
Medina et al. Gearbox fault classification using dictionary sparse based representations of vibration signals
Sina et al. Intelligent fault diagnosis of manufacturing processes using extra tree classification algorithm and feature selection strategies
Lee et al. Efficient visibility algorithm for high-frequency time-series: application to fault diagnosis with graph convolutional network
CN116910574A (zh) 基于元学习和时间卷积网络的机械设备诊断方法及装置
Zou et al. A case-learning-based paradigm for quantitative recommendation of fault diagnosis algorithms: A case study of gearbox
Guan et al. Fault diagnosis of rolling bearing with imbalanced small sample scenarios
CN113889274A (zh) 一种孤独症谱系障碍的风险预测模型构建方法及装置
GhoshRoy et al. Leveraging sampling schemes on skewed class distribution to enhance male fertility detection with ensemble AI learners
Dilbaz et al. Predictive Maintenance on Industrial Data Using Soft Voted Ensemble Classifiers