Special equipment critical process quality evaluation and early warning method based on fusion algorithm

Abstract

A special equipment key and important process quality evaluation and early warning method based on a fusion algorithm comprises the following steps: 1) collecting assembly quality data information of a special equipment general assembly workshop in real time according to a sensor deployed in the special equipment general assembly workshop; 2) constructing a multi-dimensional evaluation system for the quality of the key process of the general assembly of the special equipment, and determining the relative weight of each evaluation index in the multi-dimensional evaluation system; and 3) evaluating the assembly quality data information based on the multi-dimensional evaluation system, and obtaining an evaluation value of the quality of the special equipment final assembly key process according to the relative weight of each evaluation index. And 4) constructing a special equipment general assembly key-key process quality early warning model. And 5) inputting the evaluation value of the quality of the special equipment final assembly key-key process into the special equipment final assembly key-key process quality early-warning model to obtain the maximum posterior probability, and further determining the assembly quality early-warning grade of the special equipment final assembly workshop. The invention provides a scientific and efficient scheme for prediction and early warning of the process quality of the special equipment general assembly workshop.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, and particularly relates to a special equipment key process quality evaluation and early warning method based on a fusion algorithm. BACKGROUND

[0002] The special equipment assembly key process is essentially an assembly manufacturing process in which an operator performs fine intelligent assembly of special parts and components in a special equipment assembly intelligent workshop environment to form a special equipment finished product. Due to differences in processes, assembly equipment, and assembly component quality, the product quality assembled in the assembly workshop is not the same. Therefore, the influencing factors of the special equipment assembly workshop key process quality are complex and multi-source.

[0003] In today's enterprises, special equipment is known for its complexity of structure, diversity of components, and high standards of assembly process. According to statistics, the average number of components per type of equipment exceeds 13,000, and the hierarchical structure of components is as many as 16 layers. In the inspection link of assembly and adjustment, the user representative needs to accept 197 key process items, and each acceptance item covers 5 to 20 different specific acceptance indicators. In the face of such a large amount of data and complex processes, how to implement fine quality control has become a key problem that needs to be solved in the special equipment assembly workshop.

[0004] Around the special equipment assembly workshop key process, the factors affecting the quality of the special equipment assembly key process are divided into six categories: assembly equipment resources, operators, workshop environment, process flow, working conditions, and raw materials.

[0005] 1) Assembly equipment resources (Ae, Assembly equipment)

[0006] Assembly equipment resources refer to devices, positioning devices, holding devices, and measuring devices involved in the assembly of special equipment components. Different types of special equipment assembly tasks require different equipment resources. Even for the same assembly task, due to differences in assembly operator skills, preferences, and experience, there are certain differences in the required assembly resources. Assembly equipment resources include assembly equipment (equipment model, equipment failure rate, equipment maintenance status, equipment total running time, equipment total assembly time, equipment assembly time, etc.), positioning devices (positioning device model, positioning device maintenance status, positioning device wear), clamping devices (clamping device calibration time), and measuring devices.

[0007] 2) Operator (O, Operator)

[0008] The operator is the most important and complex uncertain factor in the process of special equipment assembly. For example, the ability of the operator to assemble the equipment, the experience of the operator, the responsible attitude of the operator, the ability of the operator to balance the tolerance, and other details may result in a large quality difference in the final special equipment product. Therefore, when considering the influence of the operator on the quality of the critical process of special assembly, the operation technology, professional quality, and personnel state of the operator should be considered.

[0009] 3) Workshop environment (E, Environment)

[0010] The quality of the critical process of special equipment assembly is easily affected by the assembly workshop environment (including the temperature, humidity, dust, vibration, and noise of the running assembly workshop). In particular, the installation of some electronic components of special equipment needs to be carried out under constant temperature conditions to minimize the reduction of the performance of electronic components. Therefore, ensuring a good and stable workshop environment is an advance to ensure the uniformity of the assembly quality of special equipment.

[0011] 4) Process flow (P, Process)

[0012] The selection of the assembly process flow largely determines the distribution of the influence of the assembly error of special equipment, such as the assembly tolerance distribution and the different selection of the geometric parameters of the equipment, which will cause the vibration of the parts of special equipment during the assembly process, thereby affecting the assembly precision. Therefore, the historical assembly data of special equipment should be considered to evaluate the influence of the assembly process flow on the quality of special equipment. The historical assembly data to be collected mainly include: the number of unqualified products under the assembly process flow, the total number of assembly, the total number of unqualified products of the same type, and the total number of assembly of the same type. In order to more effectively evaluate the influence of the assembly process flow on the assembly quality of special equipment, three quantitative evaluation indexes are defined, i.e. process unqualified rate, product unqualified rate, and process failure contribution rate. The process unqualified rate refers to the ratio of the number of unqualified products to the total number of assembly under the process flow. The product unqualified rate can be expressed as the ratio of the total number of unqualified products of the same type to the total number of assembly of special equipment. The process failure contribution rate refers to the ratio of the number of unqualified products to the total number of unqualified products of the same type under the process flow.

[0013] 5) State (S, State)

[0014] The state refers to the parameters that reflect the assembly state of the equipment in the process of special equipment assembly. Mainly include: assembly equipment parameters, clamping equipment parameters, positioning equipment parameters, and measuring equipment.

[0015] 6) Raw material (Rm, Raw material)

[0016] The raw material refers to the parts before entering the special equipment general assembly critical process, and the quality of the parts will have a superposition effect on the special equipment general assembly critical process. SUMMARY

[0017] The purpose of the present application is to provide a special equipment critical process quality evaluation and early warning method based on fusion algorithm, comprising the following steps:

[0018] 1) According to the sensors deployed in the special equipment assembly workshop, real-time acquisition of assembly quality data information of the special equipment assembly workshop.

[0019] 2) According to the fuzzy theory, a multi-dimensional evaluation system of special equipment assembly critical process quality is constructed, and the relative weights of each evaluation index in the multi-dimensional evaluation system are determined.

[0020] 3) Based on the multi-dimensional evaluation system, the assembly quality data information is judged to obtain the evaluation value of each evaluation index.

[0021] According to the relative weight of each evaluation index, the evaluation value of each evaluation index is processed to obtain the evaluation value of the special equipment assembly critical process quality.

[0022] 4) Constructing a special equipment assembly critical process quality early warning model.

[0023] 5) The evaluation value of the special equipment assembly critical process quality is input into the special equipment assembly critical process quality early warning model to obtain the maximum a posteriori probability of the quality problem in the assembly process of the special equipment assembly workshop in the critical process, and then determine the assembly quality early warning level of the special equipment assembly workshop.

[0024] Further, the evaluation indexes in the multi-dimensional evaluation system include assembly equipment resources, operating personnel, workshop environment, process flow, working condition and raw materials.

[0025] Further, in step 2), the step of determining the relative weight of each evaluation index in the multi-dimensional evaluation system comprises:

[0026] 2.1) Constructing an index relative weight judgment matrix, as shown below:

[0027] V=(v dx ,v dx ,…,v xy ,...,v xy ) (1)

[0028] In the formula, x is the index of the evaluation index, and n is the total number of evaluation indexes. V represents the relative weight judgment matrix of the current layer index. v xy is the relative weight of all evaluation indexes in the current layer to the xth evaluation index in the previous layer.

[0029] 2.2) Calculate the relative weight v of the evaluation index dx As follows:

[0030]

[0031] In the formula, The weight value parameter after row addition. The normalized weight value n xy x, y are evaluation index, n is the total number of evaluation indexes. xy The weight value of the yth evaluation index of the current layer to the xth evaluation index of the previous layer, and n xy > 0, Dx, Dy represent the xth and yth evaluation indexes respectively.

[0032] 2.3) Determine the rationality of the relative weight evaluation matrix V of the index using the random consistency ratio, and if the random consistency ratio is greater than or equal to the threshold value, adjust the consistency of the relative weight evaluation matrix V of the index.

[0033] Further, the random consistency ratio is as follows:

[0034]

[0035] In the formula, CR is the random consistency ratio. CI is the random consistency index. RI is the mean value of the random consistency index. max The maximum eigenvalue of the relative weight evaluation matrix V of the index. x is the evaluation index, and n is the total number of evaluation indexes. dx The relative weight of the evaluation index. N represents the evaluation index matrix. V represents the relative weight evaluation matrix of the current layer index.

[0036] Further, in step 3), the step of obtaining the evaluation value of the quality of the special equipment general assembly critical process includes:

[0037] 3.1) Define the quality problem evaluation set, as follows:

[0038] δ = (δ1, δ2, …, δ d , δ t ) (8)

[0039] In the formula, δ represents the quality problem evaluation set, d is the evaluation level index, t is the total number of evaluation levels, and δ d represents the quantitative value of the evaluation level of the quality problem of the special equipment general assembly critical process.

[0040] 3.2) Based on the multi-dimensional evaluation system, the assembly quality data information is evaluated, and the evaluation matrix N of each layer evaluation index is constructedi As shown below:

[0041] N i = [R1R2...R x ...R n ] T (9)

[0042] In the formula, x is the evaluation index index, n is the total number of evaluation indexes. R x represents the evaluation result of the xth evaluation index.

[0043] 3.3) Use the evaluation matrix N i of each layer evaluation index to make fuzzy comprehensive evaluation on the quality problem of special equipment assembly critical process, and get the participation degree relative vector M of each evaluation grade, as shown below:

[0044] M = ωN1 = (m1, m2, …, m n ) (10)

[0045] In the formula, ω is the relative weight vector of the previous layer evaluation index, m1-m n are participation degree vectors.

[0046] 3.4) Based on the participation degree relative vector M of each evaluation grade, the evaluation value of the quality of special equipment assembly critical process is calculated

[0047] Further, the evaluation value of the quality of special equipment assembly critical process is as follows:

[0048]

[0049] In the formula, δ represents the quality problem evaluation set, δ d represents the quantitative value of the evaluation grade of the quality problem of special equipment assembly critical process. M is the participation degree relative vector. δ T is the transpose of the quality problem evaluation set δ.

[0050] Further, the special equipment assembly critical process quality warning model is as follows:

[0051] QPM = <H(N, E), P> (12)

[0052] In the formula, QPM is the special equipment assembly critical process quality warning model. H(N, E) represents the directed acyclic graph of the quality problem of special equipment assembly critical process. N represents the parameter node vector set related to the quality problem of special equipment assembly critical process. E represents the directed edge set between two different parameter nodes with correlation. P represents the conditional probability table related to the parameter node and the correlation strength between different parameter nodes of the directed acyclic graph.

[0053] Furthermore, in step 4), the EM algorithm is used to learn the parameters of the quality early warning model for key processes in the final assembly of special equipment. The steps are as follows:

[0054] 4.1) Let N be the evaluation matrix of each level of evaluation index. i As a node variable, the total number of values ​​that the parent node can take is set to s0, and the dataset D... s ={D 1 D 2 ,…,D k ,...,D l}, where k is the node variable N i The index of the value, l is the node variable N. i The number of possible values ​​for D k This refers to node variable data with a value of k.

[0055] 4.2) Calculate the conditional probability F(D) of each node in the dataset. s |η) is shown below:

[0056]

[0057] In the formula, i is the index of the evaluation index, n is the total number of evaluation indicators, j is the index of the parent node value, and η is the index of the parent node value. ijk This indicates that when the parent node has the j-th value, the node variable N... i The probability value is k. η represents the probability.

[0058] When the parent node has the j-th value, the node variable N i The number of samples with value k, m ijk The results are obtained by formulas (14)-(15).

[0059]

[0060] In the formula, m ij This indicates that when the parent node has the j-th value, the node variable N... i The number of samples. η * This represents the optimal solution for probability.

[0061] 4.3) Obtain historical quality statistics for critical assembly processes in the special equipment final assembly workshop, and construct a quality network model instance set D. l ={x1,x2,…,x m}, and set the missing dataset S = {s1, s2, ..., s} m}, where x m This is instance data for a quality network model. m This represents missing data. m is the data index.

[0062] 4.4) The EM algorithm is used to learn the parameters of the quality warning model of the critical process of the special equipment assembly, and the conditional probability F of each node in the learned data set is calculated D (η), as follows:

[0063]

[0064] W m (s m )=p(s m |x m :η) (17)

[0065] where N0 is the total number of data. W m (s m ) is the probability density function of the missing data set S. p(s m |x m :η) is the probability density function calculation formula of the missing data set S. p(x m ,s m :η) is the probability density function calculation formula of the quality network model instance data.

[0066] Further, the maximum a posteriori probability of the quality problem occurring in the critical process assembly process of the special equipment assembly workshop is as follows:

[0067]

[0068] where P * is the maximum a posteriori probability. Q=q is the warning level. EV=ev is the evaluation value of the input special equipment assembly critical process quality. P(Q=q,EV=ev) is the probability that the input evaluation value is the current warning level. θ is the parameter that satisfies .

[0069] Further, the assembly quality warning level of the special equipment assembly workshop is divided into five levels of q1, q2, q3, q4 and q5, wherein q1 represents the safety level, q2 represents the basic safety level, q3 represents the transition level, q4 represents the slight warning level, and q5 represents the warning level.

[0070] When the assembly quality warning level of the special equipment assembly workshop is divided into q4 or q5, a warning is given.

[0071] The technical effect of the present application is self-evident. The present application solves the problems of imperfect special equipment assembly critical process quality evaluation system, difficult quantitative evaluation, inaccurate prediction, etc. The present application uses a special equipment assembly critical process quality model, combines fuzzy evaluation and Bayesian network, and provides a scientific and efficient solution for the prediction and warning of the process quality of the special equipment assembly workshop.

[0072] The application firstly establishes a special equipment assembly critical process quality evaluation system including total assembly equipment resources, operating personnel, workshop environment, process flow, working conditions, raw materials and other multi-dimensional special equipment assembly critical process quality evaluation system, and quantitatively evaluates each index; then on this basis, combined with fuzzy mathematics theory, a special equipment assembly workshop critical process evaluation method based on fuzzy comprehensive evaluation method is proposed, and finally a Bayesian network prediction model is constructed, and a Bayesian network solving method is proposed, so as to realize timely warning for special equipment assembly workshop critical process.

[0073] The application proposes a special equipment assembly critical process quality problem evaluation index body, constructs a special equipment assembly critical process quality warning model, studies the method suitable for special equipment assembly critical process quality problem evaluation and warning, analyzes the detailed process of special equipment assembly critical process quality feasible time value prediction, and realizes the prediction of special equipment assembly critical process quality feasible time value. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 It is a special equipment assembly critical process quality evaluation and warning implementation idea schematic diagram;

[0075] Figure 2 It is a special equipment assembly critical process quality evaluation index system schematic diagram;

[0076] Figure 3 It is a special equipment assembly critical process quality warning flow chart based on Bayesian network;

[0077] Figure 4 It is a special equipment assembly critical process quality warning diagram;

[0078] Figure 5 It is a special equipment assembly critical process quality model diagram;

[0079] Figure 6 It is a warning value and actual value comparison diagram;

[0080] Figure 7 It is a special equipment assembly critical process quality warning diagram. DETAILED DESCRIPTION

[0081] The application will be further described below in combination with examples, but should not be understood as limiting the above-mentioned subject matter of the application to the following examples. According to ordinary technical knowledge and conventional means in the art, various substitutions and changes can be made without departing from the technical idea of the application, and all should be included in the protection scope of the application.

[0082] Example 1:

[0083] Reference Figures 1 to 7A special equipment key process quality evaluation and early warning method based on a fusion algorithm, comprising the following steps:

[0084] 1) According to the sensors deployed in the special equipment assembly workshop, real-time acquisition of assembly quality data information of the special equipment assembly workshop.

[0085] 2) According to the fuzzy theory, a multi-dimensional evaluation system for the key process quality of the special equipment assembly is constructed, and the relative weights of each evaluation index in the multi-dimensional evaluation system are determined.

[0086] 3) Based on the multi-dimensional evaluation system, the assembly quality data information is judged to obtain the evaluation value of each evaluation index.

[0087] According to the relative weight of each evaluation index, the evaluation value of each evaluation index is processed to obtain the evaluation value of the key process quality of the special equipment assembly.

[0088] 4) Constructing a special equipment assembly key process quality early warning model.

[0089] 5) Input the evaluation value of the key process quality of the special equipment assembly into the special equipment assembly key process quality early warning model to obtain the maximum a posteriori probability of the quality problem in the assembly process of the special equipment assembly workshop in the key process, and then determine the assembly quality early warning level of the special equipment assembly workshop.

[0090] Example 2:

[0091] A special equipment key process quality evaluation and early warning method based on a fusion algorithm, the main technical content is seen in example 1, further, the evaluation index in the multi-dimensional evaluation system includes assembly equipment resources, operating personnel, workshop environment, process flow, working condition, raw materials.

[0092] Example 3:

[0093] A special equipment key process quality evaluation and early warning method based on a fusion algorithm, the main technical content is seen in any one of examples 1 to 2, further, in step 2), the step of determining the relative weight of each evaluation index in the multi-dimensional evaluation system comprises:

[0094] 2.1) Construct an index relative weight judgment matrix, as shown below:

[0095] V=(v d1 ,v d2 ,…,v dx ,...,v dn ) (1)

[0096] In the formula, x is the index of the evaluation index, and n is the total number of evaluation indexes. V represents the index relative weight judgment matrix of this layer. v dxThe relative weight of the xth evaluation index of the previous layer to all evaluation indexes of the current layer.

[0097] 2.2) Calculate the relative weight v of the evaluation index dx As shown below:

[0098]

[0099] In the formula, represents the weight value parameter after row-wise addition. n is the weight value after normalization processing xy x, y are evaluation index indices, and n is the total number of evaluation indexes. xy represents the weight value of the yth evaluation index of the current layer to the xth evaluation index of the previous layer, and n xy > 0, Dx, Dy represent the xth and yth evaluation indexes, respectively.

[0100] 2.3) Determine the rationality of the index relative weight evaluation matrix V using the random consistency ratio, and if the random consistency ratio is greater than or equal to the threshold value, adjust the consistency of the index relative weight evaluation matrix V.

[0101] Example 4:

[0102] A special equipment critical process quality evaluation and early warning method based on fusion algorithm, the main technical content is seen in any one of embodiments 1 to 3, further, the random consistency ratio is as shown below:

[0103]

[0104] In the formula, CR is the random consistency ratio. CI is the random consistency index. RI is the average value of the random consistency index. max is the maximum eigenvalue of the index relative weight evaluation matrix V. x is the evaluation index index, and n is the total number of evaluation indexes. dx is the relative weight of the evaluation index. N represents the evaluation index matrix. V represents the index relative weight evaluation matrix of the current layer.

[0105] Example 5:

[0106] A special equipment critical process quality evaluation and early warning method based on fusion algorithm, the main technical content is seen in any one of embodiments 1 to 4, further, in step 3), the step of obtaining the evaluation value of the total assembly critical process quality of the special equipment includes:

[0107] 3.1) Define the quality problem evaluation set, as shown below:

[0108] δ = (δ1, δ2, …, δd ..., δ t ) (8)

[0109] In the formula, δ represents the quality problem evaluation set, d is the evaluation grade index, t is the total number of evaluation grades, δ d represents the quantitative value of the evaluation grade of the quality problem of the special equipment general assembly critical process.

[0110] 3.2) Based on the multi-dimensional evaluation system, the assembly quality data information is evaluated, and the evaluation matrix N i of each layer evaluation index is constructed, as follows:

[0111] N i = [R1R2...R x ...R n ] T (9)

[0112] In the formula, x is the evaluation index index, and n is the total number of evaluation indexes. R x represents the evaluation result of the xth evaluation index.

[0113] 3.3) Using the evaluation matrix N i of each layer evaluation index, the fuzzy comprehensive evaluation of the quality problem of the special equipment general assembly critical process is carried out, and the participation degree relative vector M of each evaluation grade is obtained, as follows:

[0114] M = ωN i = (m1,m2,…,m n ) (10)

[0115] In the formula, ω is the relative weight vector of the previous layer evaluation index, and m1-m n are all participation degree vectors.

[0116] 3.4) Based on the participation degree relative vector M of each evaluation grade, the evaluation value of the quality of the special equipment general assembly critical process is calculated

[0117] Example 6:

[0118] A special equipment critical process quality evaluation and early warning method based on fusion algorithm, the main technical content is any one of examples 1 to 5, further, the evaluation value of the quality of the special equipment general assembly critical process is as follows:

[0119]

[0120] In the formula, δ represents the quality problem evaluation set, and δ d represents the quantitative value of the evaluation grade of the quality problem of the special equipment general assembly critical process. M is the participation degree relative vector. δT This is the transpose of the quality problem evaluation set δ.

[0121] Example 7:

[0122] A method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm is provided. The main technical contents are described in any one of Examples 1 to 6. Furthermore, the quality early warning model for critical processes in the final assembly of special equipment is as follows:

[0123] QPM =<H(N,E),P> (12)

[0124] In the formula, QPM is a quality early warning model for key processes in the final assembly of special equipment. H(N, E) represents a directed acyclic graph (DAG) of the quality problem in key processes of the final assembly of special equipment. N represents the set of parameter node vectors related to the quality problem in key processes of the final assembly of special equipment. E represents the set of directed edges between two different parameter nodes that are related. P represents the conditional probability table of the directed acyclic graph related to the parameter nodes and the correlation strength between different parameter nodes.

[0125] Example 8:

[0126] A method for quality assessment and early warning of key processes in special equipment based on a fusion algorithm is provided. The main technical contents are described in any one of Examples 1 to 7. Further, in step 4), the EM algorithm is used to learn the parameters of the quality early warning model for key processes in the final assembly of special equipment. The steps are as follows:

[0127] 4.1) Let N be the evaluation matrix of each level of evaluation index. i As a node variable, the total number of values ​​that the parent node can take is set to s0, and the dataset D... s ={D 1 D 2 ,…,D k ,...,D l}, where k is the node variable N i The index of the value, l is the node variable N. i The number of possible values ​​for D k This refers to node variable data with a value of k.

[0128] 4.2) Calculate the conditional probability F(D) of each node in the dataset. s |η) is shown below:

[0129]

[0130] In the formula, i is the index of the evaluation index, n is the total number of evaluation indicators, j is the index of the parent node value, and η is the index of the parent node value. ijk This indicates that when the parent node has the j-th value, the node variable N... i The probability value is k. η represents the probability.

[0131] where N is the node variable, and j is the value of the parent node. i m is the number of samples with value k ijk are calculated by equations (14)-(15).

[0132]

[0133] where m is the number of samples ij with value k. N is the node variable, and j is the value of the parent node. i η is the number of samples with value k. * is the optimal solution of the probability.

[0134] 4.3) Obtain the quality historical statistical data of the special equipment assembly workshop in the critical process assembly, and construct the quality network model instance set D l = {x1, x2, …, x m}, and set the missing data set S = {s1, s2, …, s m}, where x m is the quality network model instance data. s m is the missing data. m is the data index.

[0135] 4.4) Use the EM algorithm to learn the parameters of the special equipment assembly critical process quality early warning model, and calculate the conditional probability F D (η) of each node in the learned data set, as follows:

[0136]

[0137] W m (s m ) = p(s m | x m : η) (17)

[0138] where N0 is the total number of data. W m (s m ) is the probability density function of the missing data set S. p(s m | x m : η) is the probability density function calculation formula of the missing data set S. p(x m , s m : η) is the probability density function calculation formula of the quality network model instance data.

[0139] Example 9:

[0140] A method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm is provided. The main technical contents are described in any one of Examples 1 to 8. Furthermore, the maximum posterior probability of quality problems occurring in the critical assembly process of the special equipment assembly workshop is as follows:

[0141]

[0142] In the formula, P * Let θ be the maximum a posteriori probability. Q = q is the warning level. EV = ev is the input evaluation value of the quality of key processes in the final assembly of special equipment. P(Q = q, EV = ev) is the probability that the input evaluation value is the current warning level. θ is the value set by θ. The parameters that are valid.

[0143] Example 10:

[0144] A method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm is provided. The main technical contents are described in any one of Examples 1 to 9. Further, the assembly quality early warning level of the special equipment assembly workshop is divided into five levels: q1, q2, q3, q4, and q5. Among them, q1 represents the safety level, q2 represents the basic safety level, q3 represents the transition level, q4 represents the minor early warning level, and q5 represents the early warning level.

[0145] When the assembly quality warning level of the special equipment assembly workshop is classified as q4 or q5, an early warning is issued.

[0146] Example 11:

[0147] See Figures 1 to 7 A method for quality assessment and early warning of key processes in special equipment based on a fusion algorithm includes the following steps:

[0148] 1) Based on the sensors deployed in the special equipment assembly workshop, the assembly quality data of the special equipment assembly workshop is collected in real time.

[0149] 2) Based on fuzzy theory, a multi-dimensional evaluation system for the quality of key processes in the final assembly of special equipment is constructed, and the relative weights of each evaluation indicator in the multi-dimensional evaluation system are determined.

[0150] 3) The assembly quality data is evaluated based on a multi-dimensional evaluation system to obtain the evaluation values ​​of each evaluation indicator.

[0151] The evaluation values ​​of each evaluation indicator are processed according to their relative weights to obtain the evaluation values ​​of the key processes in the final assembly of special equipment.

[0152] 4) Construct a quality early warning model for key processes in the final assembly of special equipment.

[0153] 5) input the evaluation value of the quality of the special equipment assembly critical process into the quality early warning model of the special equipment assembly critical process, obtain the maximum posterior probability of the quality problem in the assembly process of the special equipment assembly workshop in the critical process, and further determine the assembly quality early warning level of the special equipment assembly workshop.

[0154] Embodiment 12:

[0155] A special equipment critical process quality evaluation and early warning method based on a fusion algorithm, the main technical content is seen in embodiment 11, further, the evaluation indexes in the multi-dimensional evaluation system include assembly equipment resources, operating personnel, workshop environment, process flow, working condition and raw materials.

[0156] Assembly equipment (Ae) refers to the equipment, positioning device, clamping device and measuring device involved in the assembly of special equipment parts. Different types of special equipment assembly tasks require different equipment resources. Even if the same assembly task, due to the different skills, preferences and experience of the assembly operators, the assembly resources required are also different. Assembly equipment resources include assembly equipment (equipment model, equipment failure rate, equipment maintenance status, equipment running total time, assembly equipment total time, etc.), positioning device (positioning device model, positioning device maintenance status, positioning device wear), clamping device (clamping device calibration time), measuring device. Operator (O) is the most important and complex uncertain factor in the assembly process of special equipment. For example, the assembly equipment capacity, assembly experience, responsible attitude and tolerance balancing ability of the operator may result in significant quality differences in the final special equipment product. Therefore, when considering the impact of operators on the quality of critical assembly processes, their operation skills, professionalism and personnel status should be considered. Environment (E), the quality of critical assembly processes of special equipment is easily affected by the assembly shop environment (including temperature, humidity, dust, vibration and noise in the assembly shop). In particular, the installation of some electronic components of special equipment needs to be carried out in a constant temperature environment to minimize the reduction of electronic component performance. Therefore, ensuring a good and stable workshop environment is an important factor in ensuring the uniformity of special equipment assembly quality. Process (P), the selection of assembly process flow largely determines the impact distribution of special equipment assembly errors, such as assembly tolerance distribution and different selection of equipment geometric parameters, which will affect the assembly precision of special equipment parts during the assembly process. Therefore, historical assembly data should be collected to evaluate the impact of assembly process flow on special equipment quality. The historical assembly data collected mainly includes: the number of unqualified products, the total number of assemblies, the total number of unqualified products of the same type, and the total number of assemblies of the same type. To more effectively evaluate the impact of assembly process flow on special equipment assembly quality, three quantitative evaluation indexes are defined: process unqualified rate, product unqualified rate and process failure contribution rate. The process unqualified rate refers to the ratio of the number of assembly unqualified products to the total number of special equipment assembly; the product unqualified rate can be expressed as the ratio of the total number of unqualified products of the same type to the total number of special equipment assembly; the process failure contribution rate refers to the ratio of the number of assembly unqualified products to the total number of unqualified products of the same type. State (S) refers to the parameters that reflect the assembly state of the equipment in real time during the assembly process of special equipment. It mainly includes: assembly equipment parameters, clamping device parameters, positioning device parameters and measuring devices.Raw material refers to the parts before entering the special equipment assembly critical process, and its quality will have a superposition effect on the special equipment assembly critical process.

[0157] Embodiment 13:

[0158] A special equipment critical process quality evaluation and early warning method based on a fusion algorithm, the main technical content of any one of embodiments 11 to 12, further, in step 2), the step of determining the relative weight of each evaluation index in the multi-dimensional evaluation system includes:

[0159] 2.1) Construct the index relative weight judgment matrix as follows:

[0160] V = (v d1 ,v d2 ,…,v dx ,...,v dn ) (1)

[0161] In the formula, x is the index of the evaluation index, n is the total number of evaluation indexes, and the value is 6. V represents the relative weight judgment matrix of the index at this level. v dx is the relative weight of all evaluation indexes at this level to the xth evaluation index at the previous level.

[0162] 2.2) Calculate the relative weight v dx of the evaluation index as follows:

[0163]

[0164] In the formula, represents the weight value parameter after adding by row. is the normalized weight value n xy . x and y are both evaluation index indexes, and n is the total number of evaluation indexes. n xy represents the weight value of the yth evaluation index at this level to the xth evaluation index at the previous level, and n xy > 0, Dx and Dy represent the xth and yth evaluation indexes, respectively.

[0165] Table 1 Special equipment assembly critical process quality evaluation index judgment matrix

[0166]

[0167] 2.3) Use random consistency ratio to judge the rationality of the index relative weight judgment matrix V, if the random consistency ratio is greater than or equal to 0.1, adjust the consistency of the index relative weight judgment matrix V.

[0168] Embodiment 14:

[0169] A special equipment key process quality evaluation and early warning method based on fusion algorithm, the main technical content is seen in any one of embodiments 11 to 13, further, the random consistency ratio is as follows:

[0170]

[0171] In the formula, CR is the random consistency ratio. CI is the random consistency index. RI is the mean value of the random consistency index. r max is the maximum eigenvalue of the index relative weight judgment matrix V. x is the evaluation index index, and n is the total number of evaluation indexes. v dx is the relative weight of the evaluation index. N represents the evaluation index matrix. V represents the relative weight judgment matrix of the index at this level.

[0172] Embodiment 15:

[0173] A special equipment key process quality evaluation and early warning method based on fusion algorithm, the main technical content is seen in any one of embodiments 11 to 14, further, in step 3), the step of obtaining the evaluation value of the special equipment assembly key process quality includes:

[0174] 3.1) Define the quality problem evaluation set, as follows:

[0175] δ=(δ1,δ2,…,δ d ,…,δ t ) (8)

[0176] In the formula, δ represents the quality problem evaluation set, d is the evaluation grade index, and t is the total number of evaluation grades, which is 5, δ d represents the quantitative value of the evaluation grade of the quality problem of the special equipment assembly key process, and (δ1, δ2, δ3, δ4, δ5) = (10, 8, 6, 4, 2).

[0177] 3.2) Based on the multi-dimensional evaluation system, the assembly quality data information is judged, and the judgment matrix N i of each layer evaluation index is constructed, as follows:

[0178] N i =[R1R2...R x ...R n ] T (9)

[0179] In the formula, x is the evaluation index index, and n is the total number of evaluation indexes. R x represents the judgment result of the xth evaluation index.

[0180] 3.3) Use the judgment matrix Ni The fuzzy comprehensive evaluation is performed on the quality problem of the critical process of the special equipment general assembly, and a participation degree relative vector M of each evaluation grade is obtained, as shown in the following formula:

[0181] M = ωN i = (m1, m2, …, m n ) (10)

[0182] In the formula, ω is a relative weight vector of the previous layer evaluation index, and m1-m n are participation degree vectors.

[0183] 3.4) Based on the participation degree relative vector M of each evaluation grade, the evaluation value of the quality of the critical process of the special equipment general assembly is calculated

[0184] Example 16:

[0185] A special equipment critical process quality evaluation and early warning method based on a fusion algorithm, the main technical content is seen in any one of examples 11 to 15, further, the evaluation value of the quality of the critical process of the special equipment general assembly is as shown in the following formula:

[0186]

[0187] In the formula, δ represents a quality problem evaluation set, and δ d represents a quantitative value of the evaluation grade of the quality problem of the critical process of the special equipment general assembly. M is a participation degree relative vector. δ T is the transpose of the quality problem evaluation set δ.

[0188] Example 17:

[0189] A special equipment critical process quality evaluation and early warning method based on a fusion algorithm, the main technical content is seen in any one of examples 11 to 16, further, the special equipment general assembly critical process quality early warning model is as shown in the following formula:

[0190] QPM = <H(N, E), P> (12)

[0191] In the formula, QPM is a special equipment general assembly critical process quality early warning model. H(N, E) represents a directed acyclic graph of the quality problem of the critical process of the special equipment general assembly. N represents a parameter node vector set related to the quality problem of the critical process of the special equipment general assembly. E represents a directed edge set between two different parameter nodes with an association relationship. P represents a conditional probability table related to the parameter node and the association strength between different parameter nodes of the directed acyclic graph.

[0192] Example 18:

[0193] A method for quality assessment and early warning of key processes in special equipment based on a fusion algorithm is provided. The main technical contents are described in any one of Examples 11 to 17. Further, in step 4), the EM algorithm is used to learn the parameters of the quality early warning model for key processes in the final assembly of special equipment. The steps are as follows:

[0194] 4.1) Let N be the evaluation matrix of each level of evaluation index. i As a node variable, the total number of values ​​that the parent node can take is set to s0, and the dataset D... s ={D 1 D 2 ,…,D k ,...,D l}, where k is the node variable N i The index of the value, l is the node variable N. i The number of possible values ​​for D k Let D be the node variable data with a value of k, and D k ={Ae k O k E k ,P k ,S k ,Rm k}

[0195] 4.2) Calculate the conditional probability F(D) of each node in the dataset. s |η) is shown below:

[0196]

[0197] In the formula, i is the index of the evaluation index, n is the total number of evaluation indicators, j is the index of the parent node value, and η is the index of the parent node value. ijk This indicates that when the parent node has the j-th value, the node variable N... i The probability value is k. η represents the probability.

[0198] When the parent node has the j-th value, the node variable N i The number of samples with value k, m ijk The results are obtained by formulas (14)-(15).

[0199]

[0200] In the formula, m ij This indicates that when the parent node has the j-th value, the node variable N... i The number of samples. η * This represents the optimal solution for probability.

[0201] 4.3) Obtain historical quality statistics for critical assembly processes in the special equipment final assembly workshop, and construct a quality network model instance set D. l{x1, x2, …, x m} and set the missing data set S = {s1, s2, …, s m}, wherein x m is the quality network model instance data. s m is the missing data. m is the data index.

[0202] 4.4) The EM algorithm is used for parameter learning of the quality early warning model of the critical process of the special equipment assembly, and the conditional probability F D (η) of each node in the learned data set is calculated, as follows:

[0203]

[0204] W m (s m ) = p(s m |x m : η) (17)

[0205] wherein N0 is the total number of data. W m (s m ) is the probability density function of the missing data set S. p(s m |x m : η) is the probability density function calculation formula of the missing data set S. p(x m , s m : η) is the probability density function calculation formula of the quality network model instance data.

[0206] Example 19:

[0207] A special equipment critical process quality evaluation and early warning method based on a fusion algorithm, the main technical content of which is any one of embodiments 11 to 18, further, the maximum a posteriori probability of the quality problem occurring in the critical process assembly of the special equipment assembly workshop is as follows:

[0208]

[0209] wherein P * is the maximum a posteriori probability. Q = q is the early warning level. EV = ev is the evaluation value of the input special equipment assembly critical process quality. P(Q = q, EV = ev) is the probability of the input evaluation value being the current early warning level. θ is the parameter that makes true.

[0210] Example 20:

[0211] A special equipment key process quality evaluation and early warning method based on fusion algorithm, the main technical content is seen in any one of embodiments 11 to 19, further, the assembly quality early warning grade of the special equipment final assembly workshop is divided into q1, q2, q3, q4, q5 five grades, wherein, q1 represents the safety level, q2 represents the basic safety level, q3 represents the transition level, q4 represents the slight early warning level, and q5 represents the early warning level. When the assembly quality early warning grade of the special equipment final assembly workshop is divided into q4 or q5, early warning is performed.

[0212] Embodiment 21:

[0213] Referring to Figures 1 to 7 A special equipment key process quality evaluation and early warning method based on fusion algorithm, including technical content including the following steps:

[0214] S1: Analyzing the current situation of the special equipment final assembly workshop, collecting the assembly quality data information of the final assembly workshop in real time according to the sensor network deployed in the special equipment final assembly workshop, and analyzing the influencing factors of the special equipment final assembly key process quality.

[0215] S2: According to the fuzzy theory, a multi-dimensional evaluation system of the special equipment final assembly key process quality is established, and the evaluation weight of each evaluation index is determined.

[0216] S3: Based on the fuzzy comprehensive evaluation method, the special equipment final assembly key process quality is comprehensively evaluated, and the feasible time value of the special equipment final assembly key process quality is evaluated.

[0217] S4: The Bayesian network prediction model is applied to predict the feasible time value of the special equipment final assembly key process quality, and provide early warning support for the advance prevention and control of the special equipment final assembly key process quality problem.

[0218] In step S1, the special equipment final assembly workshop key process is focused on, and the factors affecting the special equipment final assembly key process quality are divided into six categories: final assembly equipment resources, operating personnel, workshop environment, process flow, working condition, and raw materials.

[0219] In step S2, based on the theory of operations research, the method for determining the weight of the final assembly key process quality evaluation index based on AHP is studied. The specific calculation steps are as follows:

[0220] S01: The index relative weight evaluation matrix is established as shown below, and the nodes of the evaluation matrix are represented by n xy , wherein the elements of the evaluation matrix can be represented by n xy , and the physical meaning is the weight of each level evaluation index on the quality of the key process, that is: N=(η xy ) i×i , n xy> 0,

[0221] Table 1 Evaluation index matrix of the key process of special equipment assembly

[0222]

[0223] S02: Since the dimensions of each evaluation index are different, the elements of the evaluation matrix are normalized by using the normalization calculation according to the constructed evaluation matrix.

[0224]

[0225] S03: The normalized node parameters of each column in the table are added by row

[0226]

[0227] S04: The node parameters are normalized

[0228]

[0229] The obtained V = (v d1 , v d2 , v d3 , v d4 , v d5 , v d6 ) is the relative weight vector of the evaluation index node in the quality evaluation index system of the key process of special equipment assembly.

[0230] S05: Calculate the maximum eigenvalue r max of the relative weight vector V, that is:

[0231]

[0232] S06: Solve the consistency index CI of the evaluation matrix:

[0233]

[0234] In the formula, n is the dimension of the evaluation matrix.

[0235] S07: Obtain the random consistency ratio CR of the evaluation matrix

[0236]

[0237] where (RI) represents the average value of the random consistency index, which is generally a statistical reference value. According to the calculation formula, if (CR) < 0.1, the consistency of the relative weight evaluation matrix is good; otherwise, the consistency of the evaluation matrix needs to be adjusted.

[0238] In step S3, the fuzzy comprehensive evaluation method is used to judge the quality problems of the key assembly processes of special equipment. The specific process is as follows:

[0239] S01: Define the quality problem evaluation set.

[0240] L={q1~very serious, q2~more serious, q3~general abnormality, q4~mild abnormality, q5~no abnormality}

[0241] S02: Define the corresponding quantitative values for each evaluation set to solve the quality problems of the key assembly processes of special equipment, from qualitative description to quantitative numbers.

[0242] δ=(δ1, δ2, δ3, δ4, δ5)=(10, 8, 6, 4, 2)

[0243] S03: Define the evaluation matrix N of each layer of evaluation index i .

[0244]

[0245] S04: Calculate the participation degree of the quality problems of the key assembly processes of special equipment to each evaluation level in the evaluation set L, M=(m1, m2, m3, m4, m5, m6). M=ω·N

[0246] S05: Get the evaluation value of the quality of the key assembly processes of special equipment

[0247]

[0248] In step S4, the Bayesian network prediction model is used to predict the feasible time value of the quality of the key assembly processes of special equipment, and the specific process is as follows:

[0249] S01: According to the Bayesian network theory, construct the quality pre-warning model QPM (Quality Pre-warning Model for Key Assembly Processes) for the key assembly processes of special equipment.

[0250] QPM=<H(N, E), P>

[0251] S02: Use the EM algorithm (Expectation-Maximization algorithm, EM) to learn the parameters of the constructed model and prove the convergence of EM by the following formula.

[0252]

[0253] S03: In the assembly of critical processes in the final assembly of special equipment, based on the comprehensive process flow, historical records, and real-time collected and detected data, the constructed critical process quality early warning model is used to conduct time-segmented, multi-round Bayesian network inference on quality problems in critical processes of special equipment final assembly, and according to the maximum a posteriori probability hypothesis P... * A quality early warning chart for key processes in the final assembly of special equipment is created in chronological order, as shown in the attached chart. Figure 4 As shown. Where P * The expression is as follows:

[0254]

[0255] Example 22:

[0256] See Figures 1 to 7 A method for quality assessment and early warning of key processes in special equipment based on fusion algorithms, comprising the following technical steps:

[0257] I. Quality Evaluation Index System for Key Processes in Special Equipment Assembly

[0258] (1) Establishment of a quality evaluation index system for key processes in final assembly

[0259] The key assembly process of special equipment is essentially the fine intelligent assembly of special parts and components by operators in the intelligent workshop environment of special equipment assembly, forming the assembly manufacturing process of special equipment products. Due to the differences in process, assembly equipment, and assembly parts quality, the quality of products assembled in the assembly workshop is not the same. Therefore, the influencing factors of the quality of the key assembly process of special equipment assembly workshop are complex and multi-source. This paper divides the factors affecting the quality of the key assembly process of special equipment assembly into six categories: assembly equipment resources, operators, workshop environment, process flow, working conditions, and raw materials. Assembly equipment resources (Ae, Assembly equipment) refer to the equipment, positioning devices, holding devices, and measuring devices involved in the assembly of special equipment parts. Different types of special equipment assembly tasks require different equipment resources. Even for the same assembly task, due to the differences in assembly operator skills, preferences, and experience, the required assembly resources also have certain differences. Assembly equipment resources include assembly equipment (equipment model, equipment failure rate, equipment maintenance status, equipment running total time, equipment assembly total time, equipment assembly time, etc.), positioning devices (positioning device model, positioning device maintenance status, positioning device wear), clamping devices (clamping device calibration time), and measuring devices. Operators (O, Operator) are the most important and complex uncertain factors in the assembly process of special equipment. For example, the differences in assembly equipment capacity, assembly experience, responsible attitude, and tolerance balancing ability of operators can result in significant quality differences in the final special equipment products. Therefore, when considering the impact of operators on the quality of the key assembly process of special equipment, their operation skills, professionalism, and personnel status should be considered. Workshop environment (E, Environment): The quality of the key assembly process of special equipment is easily affected by the assembly workshop environment (including temperature, humidity, dust, vibration, and noise in the assembly workshop). In particular, the installation of some electronic components of special equipment needs to be carried out in a constant temperature environment to minimize the reduction of electronic component performance. Therefore, ensuring a good and stable workshop environment is essential to ensure the uniformity of special equipment assembly quality. Process flow (P, Process): The selection of assembly process flow largely determines the impact distribution of special equipment assembly errors, such as assembly tolerance allocation and different selection of equipment geometric parameters, which can cause vibration of special equipment parts during assembly process, thereby affecting assembly precision. Therefore, historical assembly data should be collected to evaluate the impact of assembly process flow on special equipment quality. The historical assembly data to be collected mainly include: the number of unqualified products, the total number of assemblies, the total number of unqualified products of the same type, and the total number of assemblies of the same type.To evaluate the assembly process flow on the quality of special equipment more effectively, three quantitative evaluation indexes are defined, including process unqualified rate, product unqualified rate and process failure contribution rate. The process unqualified rate refers to the ratio of the total assembly unqualified product quantity to the total assembly quantity of the special equipment. The product unqualified rate refers to the ratio of the total unqualified quantity of the same type of special equipment to the total assembly quantity of the special equipment. The process failure contribution rate refers to the ratio of the total assembly unqualified product quantity to the total unqualified quantity of the same type of special equipment. State refers to the parameter reflecting the assembly state of the equipment in the total assembly process of the special equipment. It mainly includes total assembly equipment parameters, clamping equipment parameters, positioning equipment parameters and measuring equipment. Raw material refers to the parts before entering the critical assembly process of the special equipment, and the quality of the parts has a superposition effect on the critical assembly process of the special equipment.

[0260] (2) Determination of weight value of total assembly critical process quality evaluation index system

[0261] According to the above-mentioned process quality evaluation index system, the process quality evaluation indexes are all qualitative indexes at the macro level. To realize the quantification, the evaluation weights of these indexes need to be calibrated. The analytic hierarchy process can decompose multiple and multi-level indexes layer by layer, and obtain the weight of each index through the fuzzy quantification of qualitative indexes. Therefore, based on the theory of operations research, the method for determining the weight value of the total assembly critical process quality evaluation index based on the analytic hierarchy process is studied. The steps are as follows:

[0262] s1 First, the relative weight evaluation matrix of the indexes is established. The relative weight evaluation matrix refers to the weight value comparison of all evaluation indexes in the layer relative to the evaluation index in the last layer. The node of the evaluation matrix is represented by n xy , wherein the evaluation matrix element can be represented by n xy , and the physical meaning is the weight of the quality influence of each level evaluation index on the critical process, i.e.: N=(η xy ) i×i , n xy >0, The constructed evaluation matrix is shown in Table 1.

[0263] Table 1 Evaluation matrix of total assembly critical process quality evaluation index of special equipment

[0264]

[0265] The specific steps for determining the relative weight value of the evaluation indexes are as follows:

[0266] A. Since the dimensions of each evaluation index are different, the elements of the evaluation matrix are normalized according to the normalization formula (1) based on the constructed evaluation matrix.

[0267]

[0268] B. The normalized node parameters of each column in the table are added by row

[0269]

[0270] C. The node parameters are normalized

[0271]

[0272] The obtained V = (v d1 , v d2 , v d3 , v d4 , v d5 , v d6 ) is the relative weight vector of the evaluation index node in the quality evaluation index system of the critical assembly process of special equipment.

[0273] s2 According to the relative weight vector, the consistency of the relative weight vector is verified, and a reasonable judgment is made. The consistency verification process is as follows:

[0274] A. First, calculate the maximum eigenvalue r max of the relative weight vector V according to formula (4)

[0275]

[0276] B. According to formula (5), the consistency index CI of the evaluation matrix is solved:

[0277]

[0278] In the formula, n is the dimension of the evaluation matrix.

[0279] C. The random consistency ratio CR of the evaluation matrix is obtained by formula (6):

[0280]

[0281] In formula (6), (RI) represents the average value of the random consistency index, which is generally a statistical reference value. According to the calculation formula, if (CR) < 0.1, the consistency of the relative weight evaluation matrix is good; otherwise, the consistency of the evaluation matrix needs to be adjusted.

[0282] Two, the quality evaluation of critical process based on fuzzy comprehensive evaluation method

[0283] Fuzzy comprehensive evaluation was proposed by Professor Chadd in 1965, which is a kind of evaluation method combined with fuzzy set theory. It mainly uses fuzzy mathematics theory to make comprehensive quantitative evaluation on things or objects affected by multiple factors, and can be used to evaluate uncertain attributes of things. It has the characteristics of strong systematicness and clear output, and can well handle ambiguous and difficult quantitative problems. For example, Gu Yun analyzed the risk problems of subway engineering construction in view of the heavy traffic pressure of urban commuting, constructed a risk assessment evaluation system according to a specific project, divided the risk assessment level into different levels, gave the relative weight of the evaluation system at different levels, constructed an evaluation model suitable for subway engineering risk assessment, and proposed a subway engineering evaluation method based on fuzzy comprehensive evaluation. Xu Zhengjie et al. proposed a cloud model evaluation method based on fuzzy comprehensive evaluation for the fuzzy qualitative concept in railway communication system risk assessment, which is used to solve the risk assessment of temporary speed limit project of train control center. In summary, in the application process of fuzzy comprehensive evaluation method, the evaluation objects are all constrained by multiple influencing factors, which will change with the change of evaluation time or space, and are not constant. In the evaluation process, the influence of the evaluation index on the evaluation object is reflected as the object evaluation matrix, and the result of the evaluation is the relative evaluation value obtained after a series of matrix calculations. Therefore, the fuzzy comprehensive evaluation method has a certain fuzziness, which is more in line with the objective facts of the evaluation problem itself. The quality problem of special equipment assembly critical process is affected by multiple factors, including quantitative and qualitative factors. Therefore, the evaluation of the quality problem of special equipment assembly critical process should be based on a comprehensive evaluation algorithm that supports quantitative evaluation of multiple factors. Therefore, the invention uses fuzzy comprehensive evaluation method to evaluate the quality problem of special equipment assembly critical process. First of all, the quality problem evaluation set needs to be defined. The evaluation set is a fuzzy set that describes the quality problem of special equipment assembly critical process. Combined with the classification of the quality problem of special equipment assembly critical process, the evaluation set is defined as: L = {q1 ~ very serious, q2 ~ relatively serious, q3 ~ general abnormality, q4 ~ slight abnormality, q5 ~ no abnormality}; and the corresponding quantitative values of each evaluation set are defined to solve the conversion from qualitative description to quantitative number of the quality problem of special equipment assembly critical process: δ = (δ1, δ2, δ3, δ4, δ5) = (10, 8, 6, 4, 2).

[0284] On the basis of establishing the quality problem evaluation index system of special equipment assembly critical process and defining the quality problem evaluation set of special equipment assembly critical process, the evaluation matrix N i may be defined as

[0285]

[0286] Evaluation matrix N i The qualitative evaluation index can be determined by the expert evaluation method to obtain the participation relative vector, and the participation relative vector of the quantitative evaluation index can be obtained by grade conversion: according to the difference in the value range of the quantitative evaluation index, it is divided into different conversion grades, and the actual operator can obtain the participation relative vector of each conversion grade to each evaluation grade in the evaluation set according to the historical record and management method of the quality problem of the special equipment assembly critical process. Evaluation matrix N i Each data n ijk satisfies

[0287] The evaluation results of each layer index obtained by expert scoring are N = [R1 R2 R3 R4 R5 R6] T As an evaluation matrix, the fuzzy comprehensive evaluation is used to evaluate the quality problem of the special equipment assembly critical process, and the fuzzy comprehensive evaluation formula (8) is used to calculate the participation relative vector M = (m1, m2, m3, m4, m5, m6) of the quality problem of the special equipment assembly critical process to each evaluation grade in the evaluation set L.

[0288] M = ω · N (8)

[0289] ω = (ω1, ω2, ω3, ω4, ω5, ω6) is the relative weight vector of the upper layer evaluation index. Finally, the evaluation value of the quality of the special equipment assembly critical process can be obtained according to formula (9):

[0290]

[0291] According to the above fuzzy comprehensive evaluation method, the quality problem of the special equipment assembly critical process is evaluated irregularly, and the quality evaluation value of the special equipment assembly critical process is obtained, which provides support for the prediction of the quality problem of the special equipment assembly critical process.

[0292] III. Quality prediction of special equipment assembly critical process based on Bayesian network

[0293] Bayesian network is a theoretical model for solving fuzzy concept expression and inference proposed by Cozeman F G in 1985, which analyzes fuzzy concepts and obtains inference results by simulating human thinking. Today, there are many studies in the field of Bayesian network that prove its effectiveness and use. Bayesian network is composed of nodes and the connection between nodes, and its mathematical expression is T = <H(N, E), P>. H(N, E) is a directed acyclic graph, N = (N1, N2, …, N i , …, N n(The set of nodes at each level), E is the directed edge in the directed acyclic graph, and P represents each node N. i The conditional probability table. If all nodes N are known... i Given the conditional probability, then all nodes N i The joint probability value is:

[0294] P(N) = P(N1, N2, ... N) i ,…,N n )=∏nP(N i |π(N i (10)

[0295] Wherein, π(N) i ) is node N i The set of nodes of the parent node.

[0296] Construction of a quality early warning model for key processes in the final assembly of S1 special equipment

[0297] According to Bayesian network theory, the Quality Pre-warning Model for Key Assembly Processes (QPM) for special equipment final assembly can be defined as: QPM =<H(N,E),P> ,in:

[0298] H(N,E) represents a directed acyclic graph of quality problems in key processes of special equipment assembly, consisting of two parts, N and E, which is the qualitative part of the quality early warning model for key processes in special equipment assembly.

[0299] N = (N1, N2, ..., N) i ,…,N n ) represents the set of parameter node vectors related to quality issues in key processes of special equipment assembly;

[0300] E = {E ij Let {1≤i≤n, 1≤j≤n} represent the set of directed edges between two nodes with different parameters that have an association relationship. ij Indicates from node N i To node N k A directed edge;

[0301] P = {P(N)} i |π(N i )), N i ∈N} represents a directed acyclic graph with node N. i The relevant conditional probability table and different nodes N i The correlation strength between them is a quantitative part of the quality early warning model for key processes in the final assembly of special equipment. π(N i )∈{N1, N2, ..., Nn} represents the node set of the parent node of node N i . If the variable of node N i does not exist, π(N i ) is null, and P is the prior probability of the current node.

[0302] s2 early warning model parameter fitting

[0303] The parameter learning involved in the Bayesian network refers to the solution of the conditional probability of the model node of the quality sample data by converting the likelihood value. In combination with the QPM model constructed above, it is assumed that there is a data set D s ={D 1 ,D 2 ,…,D k}, D k ={Ae k ,O k ,E k ,P k ,S k ,Rm k}, the number of values of the node variable N i is set to l, and the number of values of the parent node is set to s i , then the parameter η ijk represents the probability value of the network node variable N i when the parent node takes the jth value, and the value is k. The function expression is:

[0304]

[0305] In the formula, m ijk represents the number of samples in which the network parent node takes the jth value and the variable N i takes the value k in the quality sample data. Therefore, the function expression is formula (12).

[0306]

[0307] Due to the complex running environment of the special equipment assembly workshop and the various types of metal parts, the reliability of the workshop sensor network is difficult to be guaranteed, and thus there is a risk of packet loss in the collected assembly data. Therefore, the data set applied by the QPM model is incomplete. The parameter learning accuracy of the incomplete data set under the Bayesian network by using the maximum likelihood estimation method is not high. Therefore, the EM algorithm (Expectation-Maximization algorithm, EM) is used to learn the parameters of the constructed model. It is assumed that the quality data set is D={D1,D2,…,D k}, D l ​= {x1(l), x2(l),..., x m m If there is quality data loss, i.e. S = {s1, s2,..., s i i , then there is

[0308]

[0309] Let W i (s i ) denote the probability density function of the lost data set S, then there is and W i (S i ) ≥ 0. Bring it into equation (13), then there is:

[0310]

[0311] W i (s i ) = p(s i | x i : η) (15)

[0312] Bring equation (15) into the calculation formula of fuzzy value, then there is:

[0313]

[0314] Thus, it is the M-step, and equation (16) proves the convergence of EM.

[0315] Quality early warning of s3 special equipment general assembly critical process

[0316] In this paper, the Bayesian network probability reasoning constructed by maximum posterior hypothesis is used. First of all, according to the quality control requirements of the critical process of special equipment general assembly, the early warning levels are divided into q1, q2, q3, q4 and q5, in which q1 represents the safety level, q2 represents the basic safety level, q3 represents the transition level, q4 represents the slight early warning level, and q5 represents the early warning level; Then input the known evidence (EV) into the quality early warning model of special equipment general assembly critical process, and the maximum posterior probability hypothesis P * of the quality problem of the general assembly workshop in the critical process assembly process is obtained, that is, equation (17):

[0317] P * = argmax P (Q = q | EV = ev) (17)

[0318] Combine the Bayesian network formula to transform equation (17), then there is:

[0319]

[0320] where θ is the maximum posterior probability hypothesis the value of the parameter.

Claims

1. A method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm, characterized in that, Includes the following steps: 1) Based on sensors deployed in the special equipment assembly workshop, collect assembly quality data information in the special equipment assembly workshop in real time; 2) Based on fuzzy theory, a multi-dimensional evaluation system for the quality of key processes in the final assembly of special equipment is constructed, and the relative weights of each evaluation indicator in the multi-dimensional evaluation system are determined. 3) The assembly quality data is evaluated based on a multi-dimensional evaluation system to obtain the evaluation values ​​of each evaluation indicator; The evaluation values ​​of each evaluation indicator are processed according to their relative weights to obtain the evaluation values ​​of the quality of key processes in the final assembly of special equipment. 4) Construct a quality early warning model for key processes in the final assembly of special equipment; 5) Input the evaluation value of the quality of key processes in the special equipment final assembly into the quality early warning model of key processes in the special equipment final assembly, obtain the maximum posterior probability of quality problems occurring in the special equipment final assembly workshop during the assembly of key processes, and then determine the assembly quality early warning level of the special equipment final assembly workshop.

2. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm as described in claim 1, characterized in that, The evaluation indicators in the multi-dimensional evaluation system include final assembly equipment resources, operators, workshop environment, process flow, working conditions, and raw materials.

3. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm as described in claim 1, characterized in that, In step 2), the steps for determining the relative weights of each evaluation indicator in the multi-dimensional evaluation system include: 2.1) Construct the relative weight evaluation matrix of the indicators, as shown below: V=(v d1 ,v d2 ,…,v dx ,...,v dn ) (1) In the formula, x is the index of the evaluation index, n is the total number of evaluation indicators; V represents the relative weight evaluation matrix of the indicators in this layer; v dx This represents the relative weight of all evaluation indicators in this layer with respect to the x-th evaluation indicator in the previous layer. 2.2) Calculate the relative weights v of the evaluation indicators. dx As shown below: In the formula, This represents the weight value parameter after adding rows together; The weight values ​​n after normalization xy x and y are both indexes of evaluation indicators, and n is the total number of evaluation indicators; n xy This represents the weight value of the y-th evaluation indicator in this layer to the x-th evaluation indicator in the previous layer, and n xy >0, Dx and Dy represent the x-th and y-th evaluation indicators, respectively; 2.3) Use the random consistency ratio to judge the rationality of the relative weight evaluation matrix V of the indicators. If the random consistency ratio is greater than or equal to the threshold, then adjust the consistency of the relative weight evaluation matrix V of the indicators.

4. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm as described in claim 3, characterized in that, The random consistency ratio is as follows: In the formula, CR is the random consistency ratio; CI is the random consistency index; RI is the mean of the random consistency index; r max The largest eigenvalue of the relative weight evaluation matrix V is denoted as ; x is the index of the evaluation indicator, and n is the total number of evaluation indicators; v dx represents the relative weights of the evaluation indicators; N represents the evaluation indicator matrix; V represents the relative weight evaluation matrix of the indicators in this layer.

5. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm according to claim 1, characterized in that, In step 3), the steps for obtaining the evaluation value of the quality of key processes in the final assembly of special equipment include: 3.1) Define the quality problem judgment set as follows: δ=(δ1,δ2,…,δ d ,…,d t ) (8) In the formula, δ represents the quality problem judgment set, d is the judgment level index, t is the total number of judgment levels, and δ d A quantitative value representing the assessment level of quality problems in key processes of special equipment final assembly; 3.2) Based on a multi-dimensional evaluation system, the assembly quality data is evaluated, and an evaluation matrix N for each level of evaluation indicators is constructed. i As shown below: N i =[R1R2...R x ...R n ] T (9) In the formula, x is the index of the evaluation index, and n is the total number of evaluation indicators; R x This represents the evaluation result of the x-th evaluation indicator; 3.3) Using the evaluation matrix N of the evaluation indicators at each level i Fuzzy comprehensive evaluation was performed on the quality issues of key processes in the final assembly of special equipment to obtain the relative participation vector M for each evaluation level, as shown below: M=ωN i = (m1,m2,…,m n ) (10) In the formula, ω is the relative weight vector of the evaluation index of the previous layer, and m1-m n Both are participation vectors; 3.4) Based on the relative participation vector M of each evaluation level, calculate the evaluation value of the quality of key processes in the final assembly of special equipment.

6. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm according to claim 5, characterized in that, Evaluation value of the quality of key processes in the final assembly of special equipment As shown below: In the formula, δ represents the quality problem judgment set, δ d This represents a quantitative value indicating the assessment level of quality issues in key processes of special equipment final assembly; M is the relative participation vector; δ T This is the transpose of the quality problem evaluation set δ.

7. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm as described in claim 1, characterized in that, The quality early warning model for key processes in the final assembly of special equipment is shown below: QPM =<H(N,E),P> (12) In the formula, QPM is the quality early warning model for key processes in the final assembly of special equipment; H(N,E) represents the directed acyclic graph of the quality problem of key processes in the final assembly of special equipment; N represents the set of parameter node vectors related to the quality problem of key processes in the final assembly of special equipment; E represents the set of directed edges between two different parameter nodes with an association relationship; P represents the conditional probability table of the directed acyclic graph related to the parameter nodes and the association strength between different parameter nodes.

8. The method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm according to claim 1, characterized in that, In step 4), the EM algorithm is also used to learn the parameters of the quality early warning model for key processes in the final assembly of special equipment. The steps are as follows: 4.1) Let N be the evaluation matrix of each level of evaluation index. i As a node variable, the total number of values ​​that the parent node can take is set to s0, and the dataset D... s ={D 1 D 2 ,…,D k ,...,D l }, where k is the node variable N i The index of the value, l is the node variable N. i The number of possible values ​​for D k For node variable data with a value of k; 4.2) Calculate the conditional probability F(D) of each node in the dataset. s |η) is shown below: In the formula, i is the index of the evaluation index, n is the total number of evaluation indicators; j is the index of the parent node value, and η is the index of the parent node value. ijk This indicates that when the parent node has the j-th value, the node variable N... i The probability value that takes the value k; η represents the probability. When the parent node has the j-th value, the node variable N i The number of samples with value k, m ijk Calculated using formulas (14)-(15); In the formula, m ij This indicates that when the parent node has the j-th value, the node variable N... i The number of samples; η * This is the optimal solution for probability; 4.3) Obtain historical quality statistics for critical assembly processes in the special equipment final assembly workshop, and construct a quality network model instance set D. l ={x1,x2,…,x m }, and set the missing dataset S = {s1, s2, ..., s} m }, where x m For quality network model instance data; s m For missing data; m is the data index; 4.4) The EM algorithm is used to learn the parameters of the quality early warning model for key processes in the final assembly of special equipment, and the conditional probability F of each node in the learned dataset is calculated. D (η), as shown below: W m (s m )=p(s m |x m :h) (17) In the formula, N0 represents the total number of data points; W m (s m p(s) is the probability density function of the missing dataset S; m |x m :η) is the formula for calculating the probability density function of the missing dataset S; p(x m ,s m :η) is the formula for calculating the probability density function of instance data in the quality network model.

9. A method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm as described in claim 1, characterized in that, The maximum posterior probability of quality problems occurring during critical assembly processes in the special equipment final assembly workshop is as follows: In the formula, P * Let be the maximum a posteriori probability; Q = q be the warning level; EV = ev be the input evaluation value of the quality of key processes in the final assembly of special equipment; P(Q = q, EV = ev) be the probability that the input evaluation value is the current warning level; θ is the value of the maximum a posteriori probability. The parameters that are valid.

10. A method for quality assessment and early warning of critical processes in special equipment based on a fusion algorithm as described in claim 1, characterized in that, The assembly quality early warning level of the special equipment assembly workshop is divided into five levels: q1, q2, q3, q4, and q5. Among them, q1 represents the safety level, q2 represents the basic safety level, q3 represents the transition level, q4 represents the minor early warning level, and q5 represents the early warning level. When the assembly quality warning level of the special equipment assembly workshop is classified as q4 or q5, an early warning is issued.

Images