Method for predicting deviation of product assembly based on assembly feature map using deviation transfer

Abstract

The present invention discloses a method for predicting the deviation of a product assembly by using an assembly feature map with deviation transfer. Obtain the part list of the hydraulic press assembly, determine the assembly relationship between each part and the tolerance of each part, and use the assembly relationship between parts as part features; compile part numbers and construct an adjacency matrix and edges according to the connection relationship between nodes; obtain the functional requirement matrix and node representation of the interested node according to the shortest path tolerance chain of the interested node, and establish a functional requirement vector; determine the functional requirement vector of the interested node to determine the node label; draw the assembly feature map of the hydraulic press, input it into a neural network for training and predict the deviation of the hydraulic press assembly during service. The assembly feature map of the hydraulic press established by the present invention can be used in machine learning methods to obtain the deviation of unknown surfaces, and at the same time provides a new idea for modeling the assembly feature map of other complex mechanical products.

Description

Technical Field

[0001] The present invention relates to a method for extracting a new assembly feature map and a method for predicting the deviation of a product assembly, and proposes a method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer. Background Art

[0002] Assembly feature modeling is a key link in the modern product R & D process. The quality and efficiency of assembly feature modeling directly affect the final quality of the product. Assembly feature modeling is generally achieved by establishing the deviation transfer relationship between parts on the basis of part modeling. Existing deviation transfer networks are all based on certain reasonable assumptions. For example, the transfer process between two features follows the shortest path; when there are multiple sets of assembly relationships between two parts, the transfer is carried out along the features that mainly act on the deviation transfer. In a hydraulic press assembly, the transfer law and path of the deviation between two adjacent parts are uncertain. The deviation transfer model established based on these assumptions often cannot reflect the actual coupling law of deviations, and it is difficult to predict the deviations of other nodes based on this model in actual situations, that is, there is a lack of a deviation transfer network for the actual service stage.

[0003] Since the method based on machine learning has strong generalization ability and can model highly complex non-linear problems, in recent years, the technology of using machine learning methods to obtain the surface deviation of hydraulic press assembly parts has become a research hotspot. Therefore, it is necessary to establish a feature map reflecting the assembly relationship between the features of hydraulic press assembly parts for input to the machine learning method, and to completely define the hydraulic press assembly feature map based on the deviation transfer process. Summary of the Invention

[0004] In order to solve the problems existing in the background art, the present invention proposes a method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer, which can be applied to a hydraulic press.

[0005] The technical solution adopted by the present invention is as follows:

[0006] Step 1: Obtain the part list of the hydraulic press assembly drawing of the hydraulic press assembly, determine the assembly relationship between each part in the hydraulic press assembly, as well as the form and position tolerances, dimensional tolerances, and assembly tolerances of each part, and use the assembly relationship between parts as part features;

[0007] Step 2: Construct the nodes in the assembly feature map according to the part features, assign numbers to the parts according to the part features, and construct the adjacency matrix of the assembly feature map according to the connection relationship between each node;

[0008] Step 3: Construct the edges in the assembly feature map according to the tolerance type between the two part features;

[0009] Step 4: For the nodes of interest, look up the corresponding small displacement tensors in the tolerance type table according to the tolerance types of the nodes of interest, identify the shortest path tolerance chain of the nodes of interest, calculate the functional requirement matrix [FR] and the node representation h of the nodes of interest, and correspondingly establish the functional requirement vector [FR]*;

[0010] Step 5: Determine the functional requirement vector [FR] of the nodes of interest * The magnitudes of the elements in determine the node labels;

[0011] Step 6: Through the above five steps, draw the hydraulic press assembly feature diagram based on the deviation transfer process, and input the drawn hydraulic press assembly feature diagram into the neural network for training and predicting the deviations generated by the hydraulic press assembly during service.

[0012] In specific implementation, the product is a hydraulic press, but not limited thereto.

[0013] The specific content of Step 2 is as follows:

[0014] 2.1) In the assembly feature diagram, use part features as nodes. The number of nodes of a part feature is determined by the number of tolerance types of the part feature; for each part in the hydraulic press assembly, number the part itself with a number, and add a lowercase letter number to the actual feature of the part. The number comes first and the lowercase letter comes second as the number of each node in the feature assembly diagram;

[0015] 2.2) In the feature assembly diagram, use two concentric circles to represent each node. The outer circle is a dashed circle representing the fitting derived feature D c , and the inner circle is a solid circle representing the actual feature S c ;

[0016] The actual feature is the part feature actually machined according to the idealized shape of the part;

[0017] The fitting derived feature is extracted from the actual feature according to the spatial position relationship;

[0018] The nominal derived feature is extracted from the idealized shape of the part according to the spatial position relationship.

[0019] 2.3) Construct the adjacency matrix according to the relationship between two nodes. The adjacency matrix is a square matrix, and the row and column dimensions of the adjacency matrix are both the total number of nodes;

[0020] If there is a tolerance relationship between node v i and node v j and the two nodes belong to different parts, then the value of A ij is 1;

[0021] If node v iand node v i If there is a tolerance relationship between them and the two nodes belong to the same part, then A ii has a value of 2;

[0022] If there is no tolerance relationship between node v i and node v j or i = j, then A ij has a value of 0;

[0023] A ij represents the element in the i-th row and j-th column of the adjacency matrix.

[0024] The specific content of step 3 is as follows:

[0025] 3.1) In the described assembly feature diagram, taking the tolerances between part features as edges, between two nodes of different parts of the hydraulic press assembly or between two nodes of the same part or on the same node of the same part, according to the tolerance type, the edges of the assembly feature diagram are drawn in the following three ways:

[0026] The self-reference tolerance is represented by an arrow edge starting from a solid circle connecting the same node and pointing to a dashed circle;

[0027] The mutual-reference tolerance is represented by an edge connecting two dashed circles on the same part;

[0028] The fit tolerance is represented by an edge connecting two solid circles on two different parts;

[0029] Among them, the self-reference tolerance is a constraint tolerance type formed by the actual feature and its corresponding fitted derived feature, the mutual-reference tolerance is a constraint tolerance type between two fitted derived features in the same part, and the fit tolerance is a constraint tolerance type between the actual features that cooperate with each other between two different parts.

[0030] 3.2) Each edge marks the corresponding geometric symbol g above the edge according to the tolerance type between the two nodes.

[0031] Specifically, for geometric tolerances including perpendicularity, concentricity, parallelism, position tolerance, and flatness, etc., the symbol graphics are represented by symbols consistent with the national standard, and the fit tolerance is represented by the symbol F.

[0032] The specific content of step 4 is as follows:

[0033] 4.1) Establish a global coordinate system O at the center position of the base of the assembly, and then establish a local coordinate system i at the center position of each actual feature in the assembly;

[0034] The node numbering order means sorting first according to the numerical numbering order of the part itself, and if the numerical numbers of the parts are the same, then sorting according to the lowercase letter numbering order of the actual features of the parts.

[0035] 4.2) The tolerances between each node form a tolerance chain in the order of node numbers. There are multiple tolerance chains from any node corresponding to the base to the node of interest.

[0036] 4.3) For a tolerance chain, construct the tolerance chain matrix [FEs] as follows:

[0037] [FE s = [[FE 1 ...[FE N

[0038] where the column vector FE i is the small displacement tensor corresponding to the tolerance of the i-th node, i = 1, 2,..., N. Each node obtains the corresponding small displacement tensor by referring to the table according to the tolerance type, and N is the total number of nodes.

[0039] 4.4) For a tolerance chain, construct the Jacobian matrix [J] corresponding to the tolerance chain as follows:

[0040]

[0041] where [J] FEi is the sub-Jacobian matrix of the i-th node;

[0042] The sub-Jacobian matrix [J] FEi of the i-th node is calculated as follows:

[0043]

[0044] where [R i o 3×3 = [C xi , C yi , C zi , [R i o 3×3 is the projection matrix of the three coordinate axes of the local coordinate system i in the global coordinate system O, and C xi , C yi , C zi respectively represent the direction components of the three coordinate axes x, y, z of the local coordinate system i in the global coordinate system O; [R PTi 3×3 = [C 1 , C 2 , C 3 , [R PTi 3×3 is the projection matrix of the tolerance domain of the i-th node in the local coordinate system i, and C 1 , C​​​​​2 , C 3 respectively represent the directional components of the three directions of the tolerance field of the i-th node in the local coordinate system i; [W i n 3×3 represents the spatial position change matrix of the next node n relative to the i-th node;

[0045] The tolerance field mentioned above refers to the allowable variation range of the geometric element actually measured (such as flatness, cylindricity, etc.).

[0046] Denote the local coordinate system of the next node of the i-th node as the n coordinate system, [W i n 3×3 Calculate with the following formula:

[0047]

[0048] where respectively represent the directional components of the next node n relative to the i-th node in the x, y, z directions, [dx n , dy n , dz n are respectively the x, y, z directional components of the origin of the coordinate system n relative to the origin of the coordinate system 0, [dx i , dy i , dz i are respectively the x, y, z directional components of the origin of the coordinate system i relative to the origin of the coordinate system 0.

[0049] Through the above steps, calculate and confirm the Jacobian matrix [J] of the tolerance chain.

[0050] 4.4) For the nodes of interest, find the tolerance chain matrix [FE s corresponding to the shortest tolerance chain;

[0051] There may be multiple tolerance chain paths from the nodes of the base feature to the nodes of interest, among which there is the shortest tolerance chain. A tolerance chain is a path composed of multiple edges in the feature graph, and the path with the fewest number of nodes passed from the node representing the base feature to the node of interest is considered the shortest tolerance chain.

[0052] 4.5) Calculate the functional requirement matrix [FR] of the shortest tolerance chain to the nodes of interest through the following formula:

[0053] [FR] = [J][FEs]

[0054] ​​Among them, the functional requirement matrix [FR] has a dimension of 6×2, with 6 rows and 2 columns. The absolute value of the lower bound of each row vector is equal to the absolute value of the upper bound. Each row vector in the functional requirement matrix [FR] represents the theoretical variation range of the deviation of the node of interest in the global coordinate system O.

[0055] 4.6) Each row in the functional requirement matrix [FR] represents a tolerance domain, and the maximum value of each row is taken as the upper limit of the tolerance domain; for the node of interest, the node representation h is obtained according to the functional requirement matrix [FR]:

[0056]

[0057] Among them, h is a 10×1 matrix, representing the upper limits of the 6 tolerance domains determined by the functional requirement matrix [FR]; (x, y, z) are the three-dimensional coordinates of the origin of the local coordinate system of the node of interest in the global coordinate system O, and n is the number of nodes included in the shortest tolerance chain path;

[0058] 4.7) A 6×1-dimensional functional requirement vector [FR]* = [u*, v*, w*, α*, β*, γ*] is established according to the same dimension of the functional requirement matrix [FR] T , and each row element in [FR]* corresponds to the row vector in the functional requirement matrix [FR], representing the deviation under the actual working condition. u*, v*, w*, α*, β*, γ* respectively represent the deviations under the actual working condition of each tolerance domain, and the deviations can be measured by instruments such as micrometers and laser measuring instruments. In the actual working condition, if some elements in the vector are difficult to measure, the Monte Carlo method is used to generate them.

[0059] In the said step 5, the node label definition of each node is obtained by using the following formula, and the deviation degree of the node from weak to strong is used as the node label y c :

[0060]

[0061] Among them, [p 1 , p 2 , p 3 , p 4 , p 5 , p 6 are the weights corresponding to each item. According to the convenience of measuring the tolerances to be studied under the actual working condition, if it is easy to measure, the weight of the corresponding element is set to 1, and if the element is generated by the Monte Carlo method, its weight is set as small as possible and close to 0.

[0062] The described assembly feature diagram of the hydraulic press is a tolerance network formed by connecting part features to each other through self-reference tolerances, mutual-reference tolerances, and fit tolerances, and includes a node set, an edge set, an adjacency matrix between nodes, node representations, and node labels.

[0063] Based on the information of the parts of the assembly, part features, tolerance types, and small displacement tensors of the present invention, an assembly feature diagram of the hydraulic press is established. Nodes in the feature diagram are set, the edges and adjacency matrix of the feature diagram are defined based on three different tolerance types, the node representations are defined based on a unified Jacobian tensor model, the node labels are defined based on a comprehensive evaluation method, and the deviation data is discretized, completing the establishment of the assembly feature diagram model of the hydraulic press for the service stage.

[0064] Compared with the prior art, the present invention has the following technical effects:

[0065] By establishing a feature diagram reflecting the assembly relationship between features, and giving node representations based on the existing deviation transfer model, the deviation transfer model established based on the present invention can reflect the coupling law of actual deviations. During the actual service of the machine, based on this model, the deviations of other nodes can be accurately predicted, providing a reliable basis for predicting the maintenance parts or parts with large deviations of the hydraulic press.

[0066] The method of the present invention establishes nodes, edges, adjacency matrices, node representations, and node labels of the assembly feature diagram. The present invention can be used in machine learning methods to obtain the deviations of unknown surfaces, and at the same time provides a new idea for modeling the assembly feature diagram of other complex mechanical products. Description of the Drawings

[0067] Figure 1 It is a schematic diagram of the method for predicting the deviations of the product assembly using the assembly feature diagram with deviation transfer involved in the present invention;

[0068] Figure 2 It is a schematic diagram of each node of the cross-section A-A and cross-section B-B of the four-column hydraulic press in the specific embodiment; Figure 2 (a) represents the sectional view A-A of the four-column hydraulic press obliquely cut from the centers of the two diagonal columns, and (b) represents the sectional view B-B of the four-column hydraulic press obliquely cut from the centers of the two diagonal columns of the other diagonal.

[0069] Figure 3 It is the assembly feature diagram of the four-column hydraulic press in the specific embodiment. Specific Embodiment

[0070] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0071] Taking a certain model of four-column hydraulic press of HZ Machine Tool Factory as an example below, the schematic diagram of the four-column hydraulic press is as Figure 2As shown, an assembly feature diagram of a four-column hydraulic press is established based on the mating relationships of the parts in the four-column hydraulic press. Finally, the assembly feature diagram of this type of hydraulic press is as shown in Figure 3 . The specific steps are as follows:

[0072] Step 1: When the slider moves to the lowest point, the piston rod reaches the maximum position, and at this time, the deviation accumulation is the largest. An assembly feature diagram is established based on the position relationship of the four-column hydraulic press at this time.

[0073] Step 2: Refer to the assembly part list, dimensions, and tolerances of the four-column hydraulic press. According to the specific definitions of nodes and edges, a total of 31 nodes and 47 edges are obtained.

[0074] Step 3: To derive the node representation, the global coordinate system O takes the surface of the workbench 1 as the origin, that is, node 1a (basic feature) in the feature diagram. Local reference coordinate systems are established on each relevant surface (node), and the tolerance accumulation uses the tolerance chain along the shortest path to 1a. Taking nodes 6a, 7b, and 9a as examples, they represent the upper surface of the slider 6, the upper surface of the upper crossbeam 7, and the outer surface of the piston rod 9, respectively.

[0075] Step 4: Calculate the node representation. Taking node 6a as an example to show the detailed calculation process of the node representation, the calculation processes of other node representations are similar. The deviation propagation relationship is 1a - 3a - 3b - 6c - 6f - 6a, and the Jacobian matrix and the corresponding FEs are obtained by referring to Tables 1 and 2. According to the definition, the [FR] of 6a is:

[0076] FR 6a = [J][FE] = [[-2.2375, 2.2375], [-2.4780, 2.4780], [-3.9661, 3.9661],

[0077] [-0.0028, 0.0028], [-0.0028, 0.0028], [0, 0]] T

[0078] According to the definition of the node representation, the final node representation of 6a is shown in the following formula, and all the simplified node representations are shown in Table 6:

[0079] h 6a = [2.2375, 2.4780, 3.9661, 0.0028, 0, 0, 0, 1250, 7] T

[0080] The corresponding edges, node representations, etc. of other nodes can all be calculated by the above method.

[0081] Figure 2 The respective reference numerals in [[]] are represented as:

[0082] 1a represents the upper surface of the hydraulic press base;

[0083] 2a, 3a, 4a, 5a respectively represent the mating surfaces where the four columns contact the base;

[0084] 6a represents the mating surface where the slider of the hydraulic press contacts the cylinder barrel;

[0085] 7a represents the lower surface of the upper crossbeam;

[0086] 8a represents the mating surface where the cylinder barrel contacts the piston rod;

[0087] 9a represents the mating surface where the piston rod contacts the cylinder barrel;

[0088] 2b, 3b, 4b respectively represent the mating surfaces where columns 2, 3, 4 contact the slider;

[0089] 2c, 3c, 4c respectively represent the mating surfaces where columns 2, 3, 4 contact the upper crossbeam;

[0090] 6b, 6c, 6d, 6e respectively represent the mating surfaces where the slider contacts columns 1, 2, 3, 4; 6f represents the lower surface of the slider;

[0091] 7b represents the upper surface of the upper crossbeam;

[0092] 7c, 7d, 7e, 7f respectively represent the mating surfaces where the upper crossbeam contacts columns 1, 2, 3, 4; 7g represents the mating surface where the center of the upper crossbeam contacts the cylinder barrel;

[0093] 8b represents the mating surface where the cylinder barrel contacts the center of the upper crossbeam;

[0094] 9b represents the mating surface where the piston rod contacts the center of the slider.

[0095] Table 1 Small displacement tensors of typical tolerance types

[0096]

[0097] Table 2 Jacobian matrix and tolerance chain matrix for each node connection

[0098]

[0099]

[0100] Step 5: Calculate the node labels. Considering that the 4th and 5th elements of the interested node of the hydraulic press in the case are easier to measure in engineering, set the weights [p * in 1 , p 2 , p 3 , p 4 , p 5,p 6 The value is set to [0.01, 0.01, 0.01, 1, 1, 0.01].

[0101] Taking node 6a as an example, [FR] 6a * = [-0.0932, -1.6843, 0.0544, -0.0012, -0.0007, 0] T . The [FR] of other nodes * is calculated similarly. Calculate the y of the node according to the definition c , and according to the size of y c , the node labels are divided into 5 categories as shown in Table 3.

[0102] Table 3 Node Label Categories and Corresponding Descriptions

[0103]

[0104] Step 6: Divide each node into a training set, a validation set, and a test set, draw the obtained hydraulic press assembly feature map for training the graph neural network HGAT, and predict the deviation of the missing labels of the hydraulic press assembly. The predicted values obtained from the prediction results are used as the results of the maintenance parts and deviation degrees of the hydraulic press.

Claims

1. A method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer, characterized in that: The method comprises the following steps: Step 1: Obtain the part list of the hydraulic press assembly, determine the assembly relationships between the parts in the hydraulic press assembly, as well as the form and position tolerances, dimensional tolerances, and assembly tolerances of each part, and use the assembly relationships between the parts as part features; Step 2: Construct the nodes in the assembly feature map according to the part features, assign numbers to the parts according to the part features, and construct the adjacency matrix of the assembly feature map according to the connection relationships between the nodes; Step 3: Construct the edges in the assembly feature map according to the tolerance types between the two part features; Step 4: For the nodes of interest, look up the corresponding small displacement tensor in the table according to the tolerance type of the nodes of interest, identify the shortest path tolerance chain of the nodes of interest, calculate the functional requirement matrix [FR] and the node representation h of the nodes of interest, and correspondingly establish the functional requirement vector [FR]*; Step 5: Determine the functional requirement vector [FR] of the node of interest * The size of each element in determines the node label; Step 6: Through the above five steps, draw the hydraulic press assembly feature map based on the deviation transfer process, and input the drawn hydraulic press assembly feature map into the neural network for training and predicting the deviation generated by the hydraulic press assembly during service; The specific content of Step 3 is as follows: 3.1) In the described assembly feature map, using the tolerance between part features as the edge, between two nodes of different parts or between two nodes of the same part or on the same node of the same part, draw the edges of the assembly feature map according to the tolerance type in the following three ways: The self-reference tolerance is represented by an arrow edge starting from a solid circle connecting the same node and pointing to a dashed circle; The mutual-reference tolerance is represented by an edge connecting two dashed circles on the same part; The fit tolerance is represented by an edge connecting two solid circles on different parts; 3.2) According to the tolerance type between the two nodes, mark the corresponding geometric symbol g above the edge for each edge.

2. A method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer according to claim 1, characterized in that: The specific content of Step 2 is as follows: 2.1) In the described assembly feature map, using the part features as nodes; for each part in the hydraulic press assembly, assign a digital number to the part itself, and assign a lowercase letter number to the actual feature of the part, and use the digital number first and the lowercase letter later as the number of each node in the feature assembly map; 2.2) In the feature assembly drawing, two concentric circles are used to represent each node. The outer circle is a dashed circle representing the fitted derived feature D of the node c , and the inner circle is a solid circle representing the actual feature S of the node c ; 2.3) Construct the adjacency matrix according to the relationship between two nodes. The adjacency matrix is a square matrix, and the row and column dimensions of the adjacency matrix are both the total number of nodes; If node v i and node v j have a tolerance relationship and the two nodes belong to different parts, then the value of A ij is 1; If node v i and node v i have a tolerance relationship and the two nodes belong to the same part, then the value of A ii is 2; If node v i and node v j have no tolerance relationship or i = j, then the value of A ij is 0; A ij represents the element in the \(i\)-th row and \(j\)-th column of the adjacency matrix.

3. A method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer according to claim 1, characterized in that: The specific content of Step 4 is as follows: 4.1) Establish a global coordinate system O at the center of the base of the assembly, and then establish a local coordinate system i at the center of each actual feature in the assembly; 4.2) The tolerances between the nodes form a tolerance chain in the order of the node numbers, and there are multiple tolerance chains from any node corresponding to the base to the nodes of interest; 4.3) For a tolerance chain, construct the tolerance chain matrix [FEs] as shown below: [FE s =[[FE 1 ...[FE N ​ Among them, the column vector FE i is the small displacement tensor corresponding to the tolerance of the i-th node, where i = 1, 2,..., N, and N is the total number of nodes; 4.4) For a tolerance chain, the Jacobian matrix [J] corresponding to the tolerance chain is constructed as follows: Among them, [J] FEi is the sub-Jacobian matrix of the i-th node; The sub-Jacobian matrix [J] of the i-th node FEi The calculation formula is as follows: Among them, [R i o 3×3 =[C xi ,C yi ,C zi , [R i o 3×3 is the projection matrix of the three coordinate axes of the local coordinate system i in the global coordinate system O, and C xi , C yi , C zi respectively represent the direction components of the three coordinate axes x, y, z of the local coordinate system i in the global coordinate system O; [R PTi 3×3 =[C 1 ,C 2 ,C 3 , [R PTi 3×3 is the projection matrix of the tolerance domain of the i-th node in the local coordinate system i, and C 1 , C 2 , C 3 respectively represent the direction components of the three directions of the tolerance domain of the i-th node in the local coordinate system i; [W i n 3×3 represents the spatial position change matrix of the next node n relative to the i-th node;​​​​​ Denote the local coordinate system of the next node of the $i$-th node as the $n$ coordinate system, [W i n 3×3 Calculate it using the following formula:​ wherein respectively represent the directional components of the next node n relative to the i-th node in the x, y, and z directions, [dx n , dy n , dz n are respectively the x, y, and z directional components of the origin of coordinate system n relative to the origin of coordinate system 0, [dx i , dy i , dz i are respectively the x, y, and z directional components of the origin of coordinate system i relative to the origin of coordinate system 0; 4.4) For the nodes of interest, find the tolerance chain matrix corresponding to the shortest tolerance chain [FE s ; 4.5) Calculate the functional requirement matrix [FR] of the shortest tolerance chain to the node of interest through the following formula: [FR] = [J][FEs] where the dimension of the functional requirement matrix [FR] is 6×2, and each row vector in the functional requirement matrix [FR] represents the theoretical variation range of the deviation of the node of interest in the global coordinate system O; 4.6) Each row in the functional requirement matrix [FR] represents a tolerance domain, and the maximum value of each row is taken as the upper limit of the tolerance domain; for the node of interest, the node representation h is obtained according to the functional requirement matrix [FR]; where h is a 10×1 matrix, represents the upper limits of the six tolerance zones determined by the functional requirement matrix [FR]; (x, y, z) are the three-dimensional coordinates of the origin of the local coordinate system of the node of interest in the global coordinate system O, and n is the number of nodes included in the shortest tolerance chain path; 4.7) Set up a 6×1 functional requirement vector [FR]* = [u*, v*, w*, α*, β*, γ*] according to the same dimension of the functional requirement matrix [FR]. T , each row element in [FR]* corresponds to the row vector in the functional requirement matrix [FR], representing the deviation under the actual working conditions. u*, v*, w*, α*, β*, γ* respectively represent the deviations under the actual working conditions of each tolerance domain.

4. A method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer according to claim 3, characterized in that: In step 5, the node label of each node is defined by the following formula, and the deviation degree of the node from weak to strong is used as the node label y c : Among them, [p 1 , p 2 , p 3 , p 4 , p 5 , p 6 is the weight corresponding to each item.

5. A method for predicting the deviation of a product assembly using an assembly feature map with deviation transfer according to claim 1, characterized in that: The hydraulic press assembly feature map is a tolerance network formed by connecting part features to each other through self-reference tolerances, mutual-reference tolerances, and fit tolerances, and includes a node set, an edge set, an adjacency matrix between nodes, a node representation, and a node label.

Images