A loom fault diagnosis integrated method based on semantic ontology

Abstract

The application discloses a loom fault diagnosis integrated method based on a semantic ontology, constructs a fault tree of a loom by using fault expert knowledge of the loom, constructs a loom fault semantic ontology, collects loom equipment state monitoring data, carries out signal noise reduction processing on the collected equipment state monitoring data, inputs the equipment state monitoring data after the noise reduction processing into a trained support vector machine model, obtains a fault mechanism set, and adopts a marine predator algorithm to optimize parameters of the support vector machine model during the training. Finally, according to the classes, subclasses and the cause-and-effect relationship therebetween in the loom fault semantic ontology, the occurrence probability of each fault reason when each fault mechanism in the fault mechanism set appears is calculated, and the fault reason is investigated according to the occurrence probability of each fault reason. The application realizes integrated fault diagnosis of loom fault detection, fault judgment and fault recovery.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of complex mechanical equipment fault diagnosis, and particularly relates to a loom fault diagnosis integrated method based on a semantic ontology. BACKGROUND

[0002] The textile industry is a typical labor-intensive industry, which has long been faced with many difficulties such as strong dependence on labor, great labor cost pressure and personnel management difficulties. With the vigorous development of automation technology and new generation information technology, the textile industry gradually transforms to new forms and new modes of automation production, intelligent manufacturing, collaborative research and development design, online monitoring, etc., which provides a good development opportunity for shortening the research and development cycle of the textile industry, reducing its labor cost, and improving the overall textile operation production efficiency. At the same time, in the face of complex processes, high work intensity and heterogeneous equipment of the textile automation production line, how to timely find the fault of the key textile link, locate the fault type of the textile link, and give the fault recovery solution has become a key problem for the textile industry to complete the textile operation with quality and quantity in the new forms and new modes. The loom is an extremely important key basic manufacturing equipment in the textile industry, which has the characteristics of complex mechanical structure, harsh working environment and high maintenance cost. In the textile production process, the imperfect equipment management system, the inadequate basic maintenance, and the non-standard personnel operation will increase the risk of loom equipment failure, thereby affecting the production efficiency of the textile operation, and even causing safety accidents in serious cases. Therefore, it is of great theoretical and engineering practical significance to carry out the equipment fault diagnosis technology research of the loom system, especially to propose an integrated solution suitable for loom fault detection, fault positioning and fault recovery, to ensure the safe, stable and efficient operation of the textile operation.

[0003] In recent years, by applying artificial intelligence technologies such as multi-source information fusion, data mining, and neural networks to research on loom fault diagnosis, experts and scholars both domestically and internationally have achieved certain research results in improving the performance of loom fault diagnosis, and have also provided many new ideas for the further development of loom fault diagnosis research. However, existing loom fault diagnosis systems still face some thorny problems. For example, there is insufficient research on the management of loom fault diagnosis knowledge, and low capabilities for knowledge sharing and reuse. This means that the fault detection, fault location, and fault recovery processes for each loom are still conducted independently, making it impossible for existing methods to directly provide a matching and effective fault recovery solution based on the results of a single loom's fault diagnosis. Furthermore, even looms of the same type find it difficult to share and reuse fault diagnosis knowledge from each other. With the rapid development of ontology, ontology-based knowledge representation methods can not only clearly define domain concepts and relationships, but also achieve efficient knowledge reasoning and querying with the help of semantic network technology, making ontology-based fault knowledge management methods widely concerned in academia. Therefore, organically combining ontology with existing fault diagnosis methods is expected to provide new ideas and directions for the research of integrated solutions for loom fault diagnosis. Summary of the Invention

[0004] The purpose of this application is to propose an integrated method for loom fault diagnosis based on semantic ontology, in order to solve the problems of insufficient fault knowledge management capabilities, low knowledge sharing and reuse capabilities, and low fault diagnosis efficiency in existing loom fault diagnosis methods.

[0005] To achieve the above objectives, the technical solution of this application is as follows:

[0006] An integrated method for loom fault diagnosis based on semantic ontology, comprising:

[0007] A fault tree for the loom is constructed using expert knowledge of loom failures, and a semantic ontology of loom failures is also constructed.

[0008] Collect equipment status monitoring data of the loom and perform signal noise reduction processing on the collected equipment status monitoring data;

[0009] The noise-reduced equipment status monitoring data is input into the trained support vector machine model to obtain the set of fault-identifying mechanisms. The support vector machine model uses the ocean predator algorithm to optimize the parameters.

[0010] Based on the classes, subclasses and causal relationships in the semantic ontology of loom faults, calculate the probability of occurrence of each fault cause when each fault mechanism appears in the fault mechanism set, and investigate the fault causes based on the probability of occurrence of each fault cause.

[0011] Further, the fault tree of the loom determines the loom anomaly as the top event of the fault tree; secondly, the beating-up mechanism anomaly, the weft insertion mechanism anomaly, the take-up mechanism anomaly, the let-off mechanism anomaly and the shedding mechanism anomaly are taken as the first layer intermediate events, while in the beating-up mechanism, the weft accumulator failure, the weft selection mechanism failure and the weft detector failure are taken as the second layer intermediate events, and by analogy, the weft clutch failure in the let-off mechanism, the let-off servo failure, the broken-end sensor failure and the electronic dobby failure in the shedding mechanism are taken as the second layer intermediate events; finally, the spring wear, the yarn drum misalignment and the weft breakage are taken as the basic events of the weft accumulator failure; the weft selection plate over-flow and the stepper motor damage are taken as the basic events of the weft selection mechanism failure; the sensitivity adjustment is not proper is taken as the basic event of the weft detector failure; the guide tube breakage and the worn reed dent loosening are taken as the basic events of the weft insertion mechanism failure; the take-up gear wear or breakage and the take-up motor speed being improper are taken as the basic events of the take-up mechanism anomaly; the clutch over-flow and the clutch short circuit are taken as the basic events of the weft clutch failure; the insulation layer failure and the tension sensor setting being improper are taken as the basic events of the broken-end sensor mechanism failure; the electronic dobby angle being unreasonable, the dobby magnet being damaged and the dobby lacking lubricating oil are taken as the basic events of the electronic dobby failure.

[0012] In the loom fault semantic ontology, the shedding mechanism failure, the weft insertion mechanism failure, the beating-up mechanism failure, the let-off mechanism failure and the take-up mechanism failure are taken as the classes of the fault semantic ontology, and the shedding mechanism wear, the shedding mechanism gear breakage, the guide disc wear, the sub-dobby failure, the guide tube breakage, the worn reed dent loosening, the scissors blade wear, the weft accumulator failure, the weft selection mechanism failure, the weft detector damage, the weft clutch failure, the let-off servo failure, the broken-end sensor failure, the take-up servo failure, the take-up gear damage are taken as the subclasses; the abnormal phenomenon, the abnormal feature, the abnormal position and the abnormal composition are taken as the object attributes of the fault semantic ontology, and the historical abnormal number, the occurrence prior probability and the abnormal cause-effect relationship are taken as the data attributes of the fault semantic ontology.

[0013] Further, the signal denoising processing of the collected equipment state monitoring data comprises:

[0014] initializing the overall average number, and adding a pre-designed Gaussian white noise sequence to the collected equipment state monitoring data;

[0015] decomposing the equipment state monitoring data after adding the Gaussian white noise sequence into inherent modal components;

[0016] introducing a statistical evaluation index and a corresponding evaluation threshold value, and screening effective inherent modal components from the inherent modal component set;

[0017] reconstructing the effective inherent modal components to obtain the denoised equipment state monitoring data.

[0018] Further, the occurrence probability of each fault reason when the fault mechanism appears is calculated according to the class, subclass and the causal relationship between them in the loom fault semantic ontology, including:

[0019] According to the Bayes theorem, the occurrence probability P(B w R o ) of each fault reason when the fault mechanism appears is calculated:

[0020]

[0021] Wherein, P(B w R o ) represents the percentage of the occurrence times of the wth fault reason under the oth mechanism to the total occurrence times of loom faults, P(R o ) represents the percentage of the occurrence times of the oth mechanism to the total occurrence times of loom faults, and p w represents the percentage of the occurrence times of the wth fault reason under the oth mechanism to the total occurrence times of the oth mechanism.

[0022] The loom fault diagnosis integration method based on the semantic ontology provided in the application analyzes the fault causal chain of the loom from the knowledge level, including the mode of fault, the reason of fault and the corresponding consequence, and establishes the corresponding fault knowledge ontology; the features of the multi-source monitoring information are extracted by using the signal analysis method, and the fault mechanism of the loom is accurately identified by combining the support vector machine classifier with parameter optimization; finally, the semantic mapping relationship between the mode recognition result and the fault knowledge ontology is established by using the causal model, and the optimal maintenance strategy is obtained by combining the Bayes theorem. The application fully combines the respective advantages of machine learning in monitoring data and processing and the uncertainty reasoning of the fault ontology, so as to realize the integrated fault diagnosis of loom fault detection, fault judgment and fault recovery. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 It is a flow chart of the loom fault diagnosis integration method based on the semantic ontology of the application;

[0024] Figure 2 It is an embodiment schematic diagram of the fault tree of the loom of the application;

[0025] Figure 3 It is an embodiment part schematic diagram of the loom fault semantic ontology of the application;

[0026] Figure 4 It is an embodiment schematic diagram of the semantic ontology causal reasoning of the application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0028] In one embodiment, as shown in Figure 1 a semantic ontology-based loom fault diagnosis integration method is proposed, comprising:

[0029] Step S1, the fault tree of the loom is constructed based on the fault expert knowledge of the loom, and the loom fault semantic ontology is constructed.

[0030] In this embodiment, the fault tree of the loom is constructed according to the fault expert knowledge of the loom. First, the loom anomaly is determined as the top event of the fault tree; second, the beating mechanism anomaly, the weft insertion mechanism anomaly, the take-up mechanism anomaly, the warp let-off mechanism anomaly and the shedding mechanism anomaly are taken as the first layer intermediate events, while in the beating mechanism, the weft accumulator fault, the weft selection mechanism fault and the weft detector fault are taken as the second layer intermediate events, and similarly, the weft clutch fault, the warp let-off servo fault, the warp break sensor fault in the warp let-off mechanism and the electronic dobby fault in the shedding mechanism are selected as the second layer intermediate events; finally, the spring wear, the yarn drum misalignment and the weft breakage are taken as the basic events of the weft accumulator fault; the weft selection plate overcurrent and the stepper motor damage are taken as the basic events of the weft selection mechanism fault; the improper sensitivity adjustment is taken as the basic event of the weft detector fault; the guide tube rupture and the loose reed teeth wear are taken as the basic events of the weft insertion mechanism fault; the take-up gear wear or breakage and the improper take-up motor speed are taken as the basic events of the take-up mechanism anomaly; the clutch overcurrent and the clutch short circuit are taken as the basic events of the weft clutch fault; the insulation layer failure and the improper tension sensor setting are taken as the basic events of the warp break mechanism fault; the electronic dobby angle is unreasonable, the dobby magnet is damaged, and the dobby lacks lubricating oil as the basic events of the electronic dobby fault.

[0031] The logical relationship of the events is determined by analyzing each layer of the fault tree, and the events are connected from top to bottom by applying logic gates to form an accident chain, and the fault tree is constructed. In one specific embodiment, the constructed fault tree is shown in Figure 2 .

[0032] Then the loom fault semantic ontology is constructed, as shown in Figure 3As shown, first, the classes and subclasses in the fault semantic ontology are determined by using the built fault tree and the loom fault expert knowledge, the loom opening mechanism fault, the loom weft insertion mechanism fault, the loom beating mechanism fault, the loom warp let-off mechanism fault, the loom take-up mechanism fault are defined as the classes of the fault semantic ontology, and the loom opening mechanism wear, the loom opening mechanism gear fracture, the loom guide disc wear, the loom sub-multi-armed fault, the loom bob fracture, the loom reed tooth loosening, the loom shears blade wear, the loom weft accumulator fault, the loom weft selection mechanism fault, the loom weft detector damage, the loom weft clutch fault, the loom warp let-off servo fault, the loom warp breakage sensor fault, the loom take-up servo fault, and the loom take-up gear damage are defined as the subclasses, as shown in Table 1. Secondly, the abnormal phenomenon, the abnormal feature, the abnormal position, and the abnormal composition are defined as the object attributes of the fault semantic ontology by defining the relationship between the parent class and the subclass and the features between the faults, and the historical abnormal times, the prior probability of occurrence, and the abnormal cause-effect relationship are defined as the data attributes of the fault semantic ontology, as shown in Table 2. Then, the fault semantic ontology is filled with instances, all the fault instance individuals are filled into the classes according to the attribute relationship based on the basic knowledge of the loom fault. Finally, the consistency of the constructed semantic ontology is detected by using the Pellet inference machine of Protégé, so as to exclude the logical conflicts in the fault semantic ontology.

[0033] Through the above fault semantic ontology construction, the common loom faults are fully described from the fault knowledge level, which facilitates the management and reuse of the fault knowledge in the future, and reduces the development cost of knowledge maintenance and reuse after the ontology is built.

[0034]

[0035]

[0036] Table 1

[0037]

[0038] Table 2

[0039] Step S2, collect the loom equipment state monitoring data, and perform signal noise reduction processing on the collected equipment state monitoring data.

[0040] Firstly, the collected equipment state monitoring data is grouped according to the execution period T of the loom working task.

[0041] The loom state monitoring data collected in this embodiment is as follows:

[0042]

[0043] In formula (1) represents the sensor output data contaminated by noise, z'(k)=[z'1(k), z'2(k), …, z' S (k)]T The valid information of the collected equipment status, represented by the low-frequency noise caused by the loom take-up motion and the high-frequency noise caused by the beat-up motion, is denoted as υ(k)=[υ1(k),υ2(k),…,υ S (k)] T K represents the sampling length of each set of status monitoring data, S represents the number of sensors used to sample the status of the equipment, k = 1, ..., K, s = 1, ..., S.

[0044] Furthermore, the collected multi-source sensor data is analyzed based on the execution cycle T of the loom's work tasks. After grouping and adding labels, the complete dataset is obtained as follows:

[0045]

[0046] In formula (2) This is a floor function. When the loom is operating under normal conditions, given a label y... l (k) = 1; For a loom operating under fault conditions, given label y l (k) = -1. Furthermore, to facilitate subsequent data processing and analysis, the obtained multi-source condition monitoring dataset will be uploaded to a cloud server for storage via an industrial gateway.

[0047] It should be noted that the labeling mentioned above is for obtaining the dataset used to train the model. Once the model is trained, when diagnosing the collected loom condition monitoring data, no labels are needed; this will not be elaborated upon further below.

[0048] To address the issue of numerous interference factors on the acquired signals caused by the curling motion and other movements during loom operation, this embodiment introduces the ensemble empirical mode decomposition (EEMD) algorithm for signal noise reduction, including:

[0049] Step S2.1: Initialize the overall average frequency by adding a pre-designed Gaussian white noise sequence to the collected equipment status monitoring data.

[0050] First, initialize the overall average frequency to n, and then analyze the collected signal z. l (k) Add a pre-designed Gaussian white noise sequence:

[0051]

[0052] In equation (3), β l，i (k) is the newly calculated time series signal, u l，i (k) represents the i-th addition of white noise sequence, i = 1, 2, ..., n.

[0053] Step S2.2, decompose the device condition monitoring data after adding the Gaussian white noise sequence into a set of intrinsic mode components.

[0054] Secondly, considering the characteristic that the Gaussian white noise frequency obeys uniform distribution, β l，i (k) is decomposed into:

[0055]

[0056] In formula (4), z is the jth intrinsic mode component (IMF) decomposed after the ith white noise is added, is the residual obtained by decomposition, m is the number of IMF components, j = 1, 2,..., m, and the set of IMFs is represented as

[0057] Step S2.3, introduce statistical evaluation indexes and evaluation thresholds for z

[0058] Secondly, considering that there are pseudo-components in the IMFs after decomposition, directly reconstructing signals from these pseudo-components will reduce the accuracy of fault pattern recognition. Therefore, the embodiment introduces Q corresponding statistical evaluation indexes κ q (), q = 1,..., Q, and designs corresponding evaluation thresholds to realize screening of effective IMF components that meet the set threshold from the IMF set:

[0059]

[0060] In formula (4), z represents the effective IMF component after screening.

[0061] Step S2.4, reconstruct the device condition monitoring data after noise reduction by using the effective intrinsic mode components.

[0062] Finally, the effective IMF components z are reconstructed by:

[0063]

[0064] In formula (6), the signal after noise reduction is z l (k), and the data set after noise reduction is D = {z l (k), y l (k) | l = 1,..., L}.

[0065] Step S3, input the device state monitoring data after noise reduction processing into the trained support vector machine model to obtain a set of recognized fault mechanisms, and the support vector machine model adopts the marine predators algorithm to optimize parameters.

[0066] This embodiment uses a support vector machine model (SVM) to perform pattern recognition on the collected loom signal data set, and introduces a marine predators algorithm (MPA) to optimize the parameters of the support vector machine model, so the support vector machine model of the application is also called an MPA-SVM model.

[0067] First, the data set D is divided into training samples and test samples according to a ratio of 7:3, and the parameters are initialized. In the loom fault L-dimensional sample space, there is an optimal classification hyperplane, and its expression is:

[0068] ω·Φ(z l )+b=0 (9)

[0069] In formula (7), ω is the normal vector of the hyperplane, b is the displacement term, and Φ(·) is the mapping function. In order to obtain the estimated values of ω and b, an inequality-constrained quadratic programming problem equivalent to the optimal classification hyperplane can be obtained, and its expression is:

[0070]

[0071] In formula (8), C represents a penalty factor, which controls the complexity of the model and the approximation error, ξ l represents a relaxation variable, ε represents a small positive quantity greater than zero, y l is a data label, a Lagrange function is constructed, a kernel function is selected as a Gaussian radial basis function according to experience, and the classification decision function of the SVM model is obtained by derivation of ω, b and ξ, and dual transformation:

[0072]

[0073] In formula (9), χ is a Lagrange multiplier and is related to the penalty factor, sgn is a sign function, z is data in the data set to be judged, and the classification decision of the data is performed by calculating the size of y l .

[0074] Here, the kernel function is selected as a Gaussian radial basis function. The kernel function is to map the data from a low-dimensional space to a high-dimensional space, and the Gaussian radial kernel function is:

[0075]

[0076] Wherein, g is a parameter in the kernel function.

[0077] Since the penalty factor C in the decision function and the parameter g in the kernel function have great influence on the optimal training model of loom fault, the marine predator algorithm is used to optimize the model parameters C and g.

[0078] Define the number of iterations I, prey matrix P I and elite matrix E I are as follows:

[0079]

[0080] In formula (10), A is the size of the population, H is the dimension determined by the number of optimization parameters, a = 1, 2, …, A, h = 1, 2, …, H. In this model, H = 2, u a，h respectively represent the position of the prey in the whole space, the initial random prey matrix P I , the elite matrix E I is the position matrix when the prey matrix is copied optimally, and the iteration formula of the algorithm is:

[0081]

[0082]

[0083] In formula (11), R B is a random vector with normal distribution. is the multiplication symbol of the corresponding elements of the matrix, I max is the maximum number of iterations, is the first third of the total number of iterations, and similarly are the first half and two-thirds of the number of iterations, respectively, is a uniform random number vector in [0, 1], s I is the step size. I is the current iteration number, R λ represents a random vector subject to a power-law distribution, where is an adaptive parameter of the step size of the predator.

[0084] The obtained optimization parameter matrix γ:

[0085]

[0086] where θ is an A-dimensional row vector.

[0087] The algorithm performs global search by the prey in the first stage, i.e., the number of iterations The step size and prey matrix iteration formula are updated by formula (11); in the second stage, both the elite and the prey are looking for optimal parameters, at this time the space is divided into two parts, the first part is the number of iterations The search step length and the prey matrix iteration formula are calculated by formula (11) when the prey develops its search in the search space The search step length and the prey matrix iteration formula are calculated by formula (11) when the prey develops its search in the search space

[0088] Finally, in order to prevent falling into local optimum in the whole algorithm iteration process, the vortex and fish aggregating devices effects (FADs) are used to make the prey matrix migrate for a long distance to avoid stagnation caused by falling into local optimum, and the formula is as follows:

[0089]

[0090] In formula (13), r, γ, are random numbers in [0, 1], u' a，h , u" a，h represent random positions in the prey population, represent the maximum and minimum positions in the prey population, and the positions of part of the prey are reset by formula (13) to prevent falling into local optimum.

[0091] Finally, P I+1 is obtained through algorithm iteration, the parameters C and g are calculated by formula (12) and brought into the decision function, training classification is performed, the test set is used for verification, if it is not optimal, the above steps are repeated until the optimal MPA-SVM model is obtained.

[0092] In the embodiment, the sensor collects the loom equipment state monitoring data, and then performs signal denoising processing on the collected equipment state monitoring data. The data after denoising processing is brought into the trained model, and it is classified according to the decision function which kind of data it is, so as to judge which kind of fault mechanism data it is, and finally obtain the identified fault mechanism set: Ω = {r1, r2,..., r o ,..., r o}, wherein r o represents the result of identifying the fault mechanism, and O is the number of results.

[0093] In step S4, according to the classes, subclasses and causal relationships between them in the loom fault semantic ontology, the occurrence probability of each fault reason when each fault mechanism in the fault mechanism set appears is calculated, and the fault reasons are investigated according to the occurrence probability of each fault reason.

[0094] The classification model and the fault semantic ontology topology are combined to further deduce specific fault causes and fault recovery schemes.

[0095] As Figure 4 First, the causal strength of the fault semantic ontology knowledge and the Bayesian theorem are combined to calculate the cut set Boolean expression of each result in the recognition set as follows:

[0096] R o = B1p1∪B2p2∪...∪B W p W (14)

[0097] In formula (13), w = 1, 2, …, w, o = 1, 2, …, O, W is the maximum number of fault causes, R o represents a fault mechanism, corresponding to a class in the fault semantic ontology, B w represents a fault cause in the fault mechanism, corresponding to a subclass in the fault semantic ontology, p w represents the causal relationship between the fault cause B w and the fault mechanism R o , p w represents the percentage of the occurrence of the wth fault cause under the oth mechanism to the total number of fault occurrences of the oth mechanism.

[0098] In this embodiment, p w can be calculated according to known fault data, that is, the number of fault occurrences of the oth mechanism and the number of occurrences of the wth fault cause under the oth mechanism are counted when a fault occurs, and then p w is expressed by the percentage of the number of occurrences of the wth fault cause under the oth mechanism to the total number of fault occurrences of the oth mechanism. That is, the causal relationship is calculated by using expert experience and fault data when the semantic ontology is constructed. After the calculation, the causal relationship can be modified according to the actual calculation results.

[0099] Then, the occurrence probability P(B w |R o ) of the wth fault cause when the oth mechanism fails is calculated according to formula (14):

[0100]

[0101] wherein P(B w R o ) represents the percentage of the number of occurrences of the wth fault cause under the oth mechanism to the total number of loom faults, P(R o ) represents the percentage of the oth mechanism fault to the total number of loom faults, and p wrepresents the percentage of the number of times of the wth failure cause under the oth mechanism in the total number of times of failure of the mechanism o.

[0102] Finally, the calculated P(B w |R o ) is sorted by size, and the troubleshooting reasons are checked from high to low according to the sorting results. When repairing, the failure reasons with high probability are checked first, so that the fault can be quickly solved and production can be restored. Then, the causal relationship is modified accordingly for the next repair.

[0103] If the failure cause appears for the first time, the failure cause is added in the failure semantic ontology and Table 3. If not, the failure causal relationship p w of the corresponding class in the failure semantic ontology of the loom is updated, thereby improving the accuracy of the next failure reasoning.

[0104] The loom is repaired according to the repair measures under each failure cause as shown in Table 3. If it is a single repair measure, the repair is directly performed. If it is multiple repair measures, the Bayes theorem is used as:

[0105]

[0106] In formula (15), κ μ is the repair measure, where μ = 1, 2, … represents the number of repair measures, and P(κ μ B w ) is the percentage of the wth failure cause under the μth repair measure in the total number of repair measures. The probability of each repair measure corresponding to the failure cause is calculated by formula (15), and the repair measures are sorted according to the probability. The loom is repaired according to the order, thereby improving the repair efficiency.

[0107]

[0108] Table 3

[0109] The application introduces semantic ontology technology to improve the reuse ability and management ability of the loom failure diagnosis data;(2) fully combines the advantages of machine learning and semantic ontology to propose a mutually beneficial solution that integrates failure detection, failure judgment, and failure recovery(3) When a new failure occurs, the loom diagnosis method based on semantic ontology can record the new failure for the next reasoning, that is, the diagnostic knowledge has strong expansibility.

[0110] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific and detailed manner, but should not be construed as limiting the scope of the patent. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A semantic ontology-based integrated loom fault diagnosis method, characterized in that, The loom fault diagnosis integration method based on the semantic ontology comprises: A fault tree of the loom is constructed based on the fault expert knowledge of the loom, and a loom fault semantic ontology is constructed; Device state monitoring data of the loom is collected, and signal denoising processing is performed on the collected device state monitoring data; The device state monitoring data after the denoising processing is input into a trained support vector machine model to obtain a fault mechanism set, and the support vector machine model adopts a marine predator algorithm to optimize parameters; According to the classes, subclasses and causal relationships therebetween in the loom fault semantic ontology, the occurrence probability of each fault reason in the fault mechanism set when each fault mechanism occurs is calculated, and the fault reasons are investigated according to the occurrence probability of each fault reason. wherein the occurrence probability of each failure cause when the failure mechanism occurs is calculated according to the following formula : ； wherein, represents the percentage of the number of occurrences of the wth failure cause under the oth mechanism to the total number of occurrences of loom failures, represents the percentage of the number of occurrences of the oth mechanism failure to the total number of loom failures, represents the percentage of the number of occurrences of the wth failure cause under the oth mechanism to the total number of failures of the mechanism failure.

2. The semantic ontology-based integrated loom fault diagnosis method according to claim 1, characterized in that, The fault tree of the loom determines the loom anomaly as a top event of the fault tree; secondly, the beating-up mechanism anomaly, the weft insertion mechanism anomaly, the take-up mechanism anomaly, the warp let-off mechanism anomaly and the shedding mechanism anomaly are taken as first-layer intermediate events, and in the beating-up mechanism, the weft accumulator fault, the weft selection mechanism fault and the weft detector fault are taken as second-layer intermediate events, and by analogy, the weft clutch fault, the warp let-off servo fault, the warp breakage sensor fault in the warp let-off mechanism and the electronic dobby fault in the shedding mechanism are selected as second-layer intermediate events; finally, the spring wear, the yarn drum misalignment and the weft breakage are taken as basic events of the weft accumulator fault; The weft selection plate overcurrent and the stepping motor damage are taken as basic events of the weft selection mechanism fault; The sensitivity adjustment is taken as a basic event of the weft detector fault, the guide tube breakage and the wear reed tooth loosening are taken as basic events of the weft insertion mechanism fault; The take-up gear wear or breakage and the take-up motor speed are taken as basic events of the take-up mechanism anomaly; the clutch overcurrent and the clutch short circuit are taken as basic events of the weft clutch fault; the insulation layer failure and the tension sensor setting are taken as basic events of the warp breakage mechanism fault; the electronic dobby angle is unreasonable, the dobby magnet is damaged, and the dobby lacks lubricating oil, which are taken as basic events of the electronic dobby fault; In the loom fault semantic ontology, the shedding mechanism fault, the weft insertion mechanism fault, the beating-up mechanism fault, the warp let-off mechanism fault and the take-up mechanism fault are taken as classes of the fault semantic ontology, and the shedding mechanism wear, the shedding mechanism gear breakage, the guide disc wear, the dobby fault, the yarn tube breakage, the wear reed tooth loosening, the shear blade wear, the weft accumulator fault, the weft selection mechanism fault, the weft detector damage, the weft clutch fault, the warp let-off servo fault, the warp breakage sensor fault, the take-up servo fault, the take-up gear damage are taken as subclasses; the abnormal phenomenon, the abnormal feature, the abnormal position and the abnormal composition are taken as object attributes of the fault semantic ontology, and the historical abnormal number, the occurrence prior probability and the abnormal causal relationship are taken as data attributes of the fault semantic ontology.

3. The semantic ontology based integrated loom fault diagnosis method according to claim 1, characterized in that, The signal denoising processing on the collected device state monitoring data comprises: Initializing the total average number, adding a pre-designed Gaussian white noise sequence to the collected device state monitoring data; The device state monitoring data after the Gaussian white noise sequence is added is decomposed into inherent modal components; Introducing statistical evaluation indexes and corresponding evaluation thresholds, and screening effective intrinsic modal components from the set of intrinsic modal components; Reconstructing the device state monitoring data after noise reduction by using the effective intrinsic modal components.

Images