Online Detection Precision Evaluation Method for Machined Product Quality Based on Multi-Factor Fusion

Abstract

The present invention discloses an online detection accuracy evaluation method for machining product quality based on multi-factor fusion. This method first determines the input factor characteristics of the machining process for online detection accuracy evaluation and the output machining accuracy grade. Secondly, it analyzes the historical samples of the factor characteristics, generates an atomized template set, and constructs a random fuzzy template set. The online input feature samples are transformed into a random fuzzy set to be detected, and then the random fuzzy template set is matched with the random fuzzy set to be detected to obtain the credibility distribution describing the non-linear mapping relationship between multi-environment factor characteristics and detection accuracy. Finally, the credibility distributions are fused, and the accuracy grade of the online detection of machining product quality is determined from the fusion result to judge whether the product is qualified. The present invention can handle complex non-linear relationships and efficiently obtain the credibility distribution related to the detection accuracy, and more accurately evaluate the online detection accuracy in practical applications.

Description

Technical Field

[0001] The present invention belongs to the fields of intelligent manufacturing and quality control, and particularly relates to an online detection accuracy evaluation method for machining product quality based on multi-factor fusion. Background Art

[0002] In modern manufacturing, the machining accuracy of products directly affects their performance and service life. Especially in the field of high-precision manufacturing, the dimensional accuracy grade of machined products is a key indicator to ensure product quality. However, due to many uncertain factors in the machining process, such as the stability of machining equipment, the fluctuation of environmental temperature, the change of workpiece material properties, etc., traditional dimensional accuracy detection methods are difficult to cope with these complex working conditions, often resulting in inaccurate and unreliable detection results. Statistical data shows that the unqualified rate of products due to insufficient dimensional accuracy in production is about 10%, and the defective rate caused by detection errors accounts for a large proportion.

[0003] The intelligent machining product accuracy evaluation (calibration) method is to perform adaptive calibration on the dimensional accuracy of products according to real-time data, environmental changes and equipment operating status during the machining process. In terms of dimensional detection, this method can keep the calibration parameters consistent with the actual machining state, provide a more accurate accuracy grade judgment, and thus effectively improve the machining quality of products. Different from traditional accuracy detection methods, the online detection accuracy evaluation method for machining product quality based on multi-factor fusion can automatically adjust the calibration strategy and correct the detection accuracy in real time by perceiving the operating conditions of machining equipment (such as spindle speed, feed speed, cutting force, etc.), environmental conditions (such as temperature, humidity, etc.), material properties (such as hardness, elasticity, etc.) and changes in machining tasks (such as finish machining, rough machining, etc.). This can overcome the detection errors caused by complex working conditions or environmental changes, ensure the accuracy stability during the machining process, and guarantee product quality.

[0004] Under this background, in order to improve the quality detection accuracy of machined products and reduce the defective rate of products caused by dimensional errors, there is an urgent need for an online detection method that can calibrate the dimensional accuracy grade of machined products in real time and accurately. The proposed online detection accuracy evaluation method for machining product quality based on multi-factor fusion can combine multiple influencing factors, dynamically adjust and calibrate the dimensional changes during the machining process to ensure high-precision quality control in a complex environment. This is of great significance for improving product quality, reducing rework costs and increasing production efficiency. Summary of the Invention

[0005] In view of the above problems, the present invention proposes an online detection accuracy evaluation method for machining product quality based on multi-factor fusion. This method first constructs a random fuzzy template set through historical sample data. After the actual detection data is obtained online, a random fuzzy set to be detected is generated, and it is matched with the random fuzzy template set in real time to obtain the reliability distribution of the accuracy level. Subsequently, the reliability distribution is comprehensively analyzed through a reliability fusion model, and finally the accuracy evaluation result of the online data is given. Based on the accuracy level obtained through the evaluation, the actual measured value is calibrated, and it is compared with the upper and lower tolerance ranges of the geometric dimensions of the product to determine whether the product meets the quality standards.

[0006] The online detection accuracy evaluation method for machining product quality based on multi-factor fusion proposed by the present invention includes the following steps:

[0007] (1) Determine the input factor characteristics and output accuracy levels for multi-factor fusion in online detection accuracy evaluation;

[0008] (2) Conduct a coarse-grained analysis of historical samples, obtain the key parameters of the atomized template set, the expected value (Ex), entropy (En), and hyper-entropy (He), and generate the atomized template set;

[0009] (3) Based on the atomized template set generated in step (2), calculate the KL divergence between it and the historical sample set respectively, and further construct a random fuzzy template set;

[0010] (4) Obtain the input factor characteristic data during product quality detection online, and construct a random fuzzy set to be detected regarding the input factor characteristic data;

[0011] (5) Based on steps (3) and (4), match the random fuzzy template set with the random fuzzy set to be detected to obtain the reliability distribution describing the non-linear mapping relationship between multi-environment factor characteristics and detection accuracy;

[0012] (6) Fuse the reliability distributions of multiple environment factor characteristics, and determine the accuracy level of online product quality detection according to the fusion result;

[0013] (7) After evaluating the actual measured value using the accuracy level, compare the evaluated value with the upper and lower tolerances of the geometric dimensions measured of the product to determine whether the product is qualified.

[0014] The present invention proposes an online detection accuracy evaluation method for machining product quality based on multi-factor fusion. First, determine the multi-factor fusion model for online detection accuracy evaluation, including input factor characteristics and output accuracy levels. Through the coarse-grained analysis of historical samples, extract the key parameters of the atomization template set to generate the atomization template set. Based on this set, calculate its KL divergence from the historical sample set, and further construct a complete random fuzzy template set. After obtaining the input factor characteristic data of product quality detection online, generate a corresponding random fuzzy set to be detected. By matching the random fuzzy template set with the random fuzzy set to be detected, obtain the credibility distribution describing the non-linear mapping relationship between multi-environment factor characteristics and detection accuracy. Subsequently, fuse the credibility distributions of multiple factor characteristics to finally determine the accuracy level of online product quality detection. According to the accuracy level, evaluate the actual measurement value and compare the evaluation value with the upper and lower tolerance ranges of the product geometric dimensions to comprehensively judge whether the product meets the quality standards.

[0015] The present invention provides an online detection accuracy evaluation method for machining product quality based on multi-factor fusion, which has the following beneficial effects:

[0016] (1) Improve the accuracy of online detection accuracy evaluation. By constructing multiple input factor characteristics of the multi-factor fusion model, the present invention can accurately evaluate the online detection accuracy of machining product quality. Using the key parameter extraction technology of the atomization template set and combining the calculation of KL divergence to further construct a random fuzzy template set enables the model to handle complex non-linear relationships, thereby improving the accuracy of detection accuracy evaluation.

[0017] (2) Effectively handle the correlation between historical samples and real-time data. Through the coarse-grained analysis of historical samples and the random fuzzy template set constructed based on KL divergence, the relationship between historical sample data and real-time detection data can be effectively mapped. In real-time detection, based on the matching of the random fuzzy set to be detected and the random fuzzy template set, the credibility distribution related to detection accuracy can be efficiently obtained, so as to more accurately evaluate the online detection accuracy in practical applications.

[0018] (3) Support the fusion analysis of multi-environment factors. The present invention adopts a multi-factor fusion method, comprehensively considers multiple factors affecting detection accuracy, and through the fusion of credibility distributions, can effectively evaluate the detection accuracy under complex working conditions. This method is especially suitable for production scenarios where environmental factors change frequently, ensuring reliable detection results under diverse environmental conditions.

[0019] (4) Improve the reliability of quality assessment and judgment. By comparing with the upper and lower tolerance ranges of the product's geometric dimensions, the present invention can not only accurately assess the detection accuracy of the product quality, but also further ensure whether the product meets the quality standards. This provides a more stable and reliable basis for quality assessment in the manufacturing industry, helping enterprises to timely detect and correct product defects during the production process.

[0020] (5) Have good applicability and promotion value. The method of the present invention can be widely applied to the fields of quality control, quality management, and process optimization of machined products, especially suitable for precision manufacturing and high-requirement detection scenarios. Through the comprehensive assessment of multi-factor characteristics, the present invention has strong adaptability, can adjust the model parameters according to different process requirements, and has good generality and promotion value. Description of the Drawings

[0021] Figure 1 is a flow block diagram for online detection accuracy assessment of machined product quality based on multi-factor fusion;

[0022] Figure 2 is a structural block diagram for online detection accuracy assessment of machined product quality based on multi-factor fusion;

[0023] Figure 3 is a matching diagram of the random fuzzy template set and the random fuzzy set to be detected. Detailed Embodiment

[0024] The embodiment of the present application proposes an online detection accuracy calibration method for machined product quality based on multi-factor fusion, and its flow block diagram is as Figure 1 shown, including the following steps:

[0025] (1) Determine the input factor characteristics and the output accuracy level for multi-factor fusion of online detection accuracy assessment;

[0026] (2) Conduct a coarse-grained analysis of historical samples, obtain the key parameters of the atomized template set, the expected value (Ex), entropy (En), and hyper-entropy (He), and generate the atomized template set;

[0027] (3) Based on the atomized template set generated in step (2), calculate the KL divergence between it and the historical sample set respectively, and further construct a random fuzzy template set;

[0028] (4) Online obtain the input factor characteristic data during product quality detection, and construct a random fuzzy set to be detected about the input factor characteristic data;

[0029] (5) Based on steps (3) and (4), match the random fuzzy template set with the random fuzzy set to be detected, and obtain the belief distribution describing the non-linear mapping relationship between multi-environment factor characteristics and detection accuracy;

[0030] (6)Fuse the reliability distributions of multiple environmental factor features, and determine the accuracy level of on-line product quality inspection according to the fusion result;

[0031] (7)After evaluating the actual measurement value using the accuracy level, compare the evaluation value with the upper tolerance and lower tolerance of the measured geometric dimension of the product to determine whether the product is qualified.

[0032] The specific description of the said step (1) is as follows:

[0033] According to the geometric dimension type and processing technology that need to be detected for the machined product, determine the factor feature vector X = [x 1 ,..., x i ,...., x I (I = 5) as the input of the multi-factor fusion model for on-line inspection accuracy evaluation, which respectively represent the external environmental temperature x 1 , the external environmental humidity x 2 , the mechanical amplitude x 3 of the detection unit, the mechanical vibration frequency x 4 of the detection unit, and the temperature x 5 of the detection unit; Take the accuracy level {m 1 ,..., m d ,..., m D}(D = 3) as the output of the fusion model, which respectively represent high accuracy m 1 , medium accuracy m 2 , low accuracy m 3 . Then, the feature sample set composed of J historical samples is denoted as S data = {[X(j), m d |j = 1, 2,..., J; d = 1, 2,..., D}, and X(j) represents the j-th factor feature vector set.

[0034] For the convenience of understanding, an example is given here to illustrate step (1), and the process is as follows:

[0035] During the processing of a certain mechanical component, determine the factor feature vector X = [x 1 , x 2 , x 3 , x 4 , x 5 , which respectively represent the external environmental temperature x 1 , the external environmental humidity x 2 , the mechanical amplitude x 3 of the detection unit, the mechanical vibration frequency x 4 of the detection unit, and the temperature x 5 of the detection unit; Take the accuracy level {m1 ,m 2 ,m 3} as the output of the fusion model, representing high precision m 1 , medium precision 2 , low precision 3 In the process of machining this mechanical component, J = 180 historical samples were collected, which constitute the feature sample set denoted as S data ={[X(j),m d ]|j=1,2,...,180; d=1,2,3}.

[0036] The specific description of step (2) is as follows:

[0037] (2-1) Filter historical samples S according to the same accuracy level corresponding to the factor feature vector X data , get the factor feature x i (i=1,2,...,I), the accuracy level is m d A set of N samples (d=1,2,...,D) x i (n) represents the nth sample in the set.

[0038] (2-2) Based on the data obtained in step (2-1) According to formula (1) to formula (3), we can obtain the Atomization sample collection Key parameters (Ex, En, He):

[0039]

[0040] (2-3) With En as the expectation and He as the standard deviation, generate a normal random number y p , which satisfies Then take Ex as the expectation, |y p | is the standard deviation, generate another normal random number satisfy and The corresponding degree of certainty It can be calculated by formula (4). So far, a subset of the atomization sample set can be obtained.

[0041]

[0042] (2-4) Repeat step (3-3) P times until P normal random numbers and their corresponding degrees of certainty are generated to form a fog sample set. Where P = 5000.

[0043] For ease of understanding, the following is an example to illustrate step (2):

[0044] Filter the 180 sample data in the historical sample S data ={[X(j),m d |j = 1, 2,..., 180; d = 1, 2, 3} to obtain data sets with precision levels of m 1 、m 2 、m 3 under different factor characteristics According to step (2-4), the corresponding atomization template set can be obtained The specific description of step (3) is as follows:

[0045]

[0046] (3-1) Based on the atomization template set obtained in step (2-4) Arrange the P normal random numbers it contains in ascending order of their values to obtain the sorted set of normal random numbers Satisfy

[0047] (3-2) Evenly divide the sorted set of normal random numbers into two parts, namely the low-order subset and the high-order subset where

[0048] (3-3) Similarly, the N sample sets in can successively obtain the low-order subset and the high-order subset based on steps (3-1) and (3-2). If N is odd, assign the middle element to either group

[0049] (3-4) Based on the low-order subset and the high-order subset obtained in step (3-2), calculate their means μ L 、μ H and standard deviations σ L 、σ H , and calculate the KL divergence value through Equation (5)

[0050]

[0051] At the same time, the means μ' of the low-order subset and the high-order subset can be obtained​L , μ' H Standard deviation σ' L , σ′ H , further calculate and KL divergence value of

[0052]

[0053] Finally, judge by Equation (7) atomization degree of

[0054]

[0055] (3 - 5) Randomly and with replacement group the N samples in to obtain a total of S groups of data with n samples in each group s , where the s-th group of data can be expressed as where S and n s are obtained from Equations (8) and (9):

[0056] S = 1 + log 2 (P) (8)

[0057]

[0058] where P represents the number of normal random numbers in step (3 - 4), and P ≥ S.

[0059] (3 - 6) For the s-th group of data T s , calculate its expected value Ex s and standard deviation σ s :

[0060]

[0061] Similarly, the means (Ex 1 , Ex 2 ,..., Ex s ,..., Ex S ) and standard deviations (σ 1 , σ 2 ,..., σ s ,..., σ S ) of all S groups of data can be obtained.

[0062] (3 - 7) According to the atomization degree estimation in step (3 - 4), if is judged to be in a non-atomized state, obtain the key parameters of the random fuzzy template set through Equations (12) to (14)

[0063]

[0064]

[0065] where q 1 = max{Ex 1 + 3σ 1 ,..., Ex S + 3σ S}, q 2 = max{Ex 1 - 3σ 1 ,..., Ex S - 3σ S}, q 3 = min{Ex 1 + 3σ 1 ,..., Ex S + 3σ S}, q 4 = min{Ex 1 - 3σ 1 ,..., Ex S - 3σ S};

[0066] If is judged to be in the atomization state, the key parameters of the random fuzzy template set are obtained through Equations (15) to (18)

[0067]

[0068] where represents the sample variance of and represents

[0069] (3 - 8) Based on the method in step (3 - 7), the key parameters of the random fuzzy template set with the accuracy level of m i under the factor feature x d can be obtained The corresponding key parameters For the 3 accuracy levels under 5 factor features, 15 key parameters of the corresponding random fuzzy template sets can be obtained, as shown in Table 1:

[0070] Table 1 Key parameters of the random fuzzy template set of

[0071]

[0072] (3 - 9) Based on the key parameters of the random fuzzy template set obtained in step (3 - 8) Substituting Ex, En and He in step (2-3) in turn, according to steps (2-3) and (2-4), we can obtain 15 random fuzzy sample sets corresponding to 3 accuracy levels under 5 factor characteristics.

[0073] In the external environment temperature characteristic x 1 The random fuzzy sample sets of three levels of accuracy are

[0074] In the external environment humidity characteristics x 2 Below, a set of random fuzzy templates with three levels of accuracy

[0075] In the detection unit mechanical amplitude characteristic x 3 The random fuzzy sample sets of three levels of accuracy are

[0076] In the detection unit mechanical vibration frequency characteristic x 4 Below, a set of random fuzzy templates with three levels of accuracy

[0077] In the detection unit temperature characteristic x 5 Below, a set of random fuzzy templates with three levels of accuracy

[0078] For ease of understanding, step (3) is illustrated here as an example. The process is as follows:

[0079] Calculate the historical sample screening The KL divergence value of The KL divergence value is compared to judge The atomization state is further obtained The key parameters of the random fuzzy template set As shown in Table 2:

[0080] Table 2 Random fuzzy sample set Key parameters

[0081]

[0082] Based on the key parameters of the random fuzzy sample set in Table 2, 15 random fuzzy sample sets can be obtained:

[0083] The specific description of step (4) is as follows:

[0084] (4-1) When obtaining a set of data to be evaluated online The key parameter Ex of the random fuzzy set to be inspected can be calculated by Equation (19) and En is determined based on the key parameter of the random fuzzy template set i and He i : i

[0085]

[0086] (4-2) Based on the key parameters (Ex i ,, En i , He i ) of the random fuzzy set to be inspected obtained in step (4-1), successively replace Ex, En, and He in step (2-3). According to steps (2-3) and (2-4), the random fuzzy set to be inspected under 5 factor characteristics can be obtained

[0087] For easy understanding, here an example is given to illustrate the acquisition process of the random fuzzy set to be inspected when the online evaluation data is The process is as follows:

[0088] Obtain the key parameters of the random fuzzy set to be inspected according to Equations (19) to (21), as shown in Table 3

[0089] Table 3 Key parameters of the random fuzzy set to be inspected

[0090]

[0091] Based on the key parameters, the random fuzzy set to be inspected can be obtained

[0092] The specific description of step (5) is as follows:

[0093] (5-1) Construct the expected curve and entropy-containing expected curve of the random fuzzy template set which are and The expected curve and entropy-containing expected curve of the random fuzzy set to be inspected are and

[0094] (5-2) Under the factor characteristic x i , obtain the boundary values of the intersection region of the random fuzzy template set and the random fuzzy set to be inspected at 3 precision levels. The specific calculation method is as follows:​​​

[0095] Calculate and the abscissa value of the intersection point as well as the curve and V i the abscissa value of the intersection point At the same time, obtain the curve and V i the ordinate value of the intersection point as well as the curve and the ordinate value of the intersection point By obtaining the above four points, the random fuzzy template set and the random fuzzy set to be inspected the subset range of the intersection area can be determined.

[0096] (5-3) According to the range determined in step (5-2), select H subset elements that satisfy equation (22) from the random fuzzy template set while selecting h subset elements c that satisfy equation (23) from the random fuzzy set to be inspected

[0097]

[0098] At the same time, from the random fuzzy set to be inspected in which satisfy equation (23) h subset elements c i (p):

[0099]

[0100] Finally, according to equation (24), the matching value of the random fuzzy template set i and the random fuzzy set to be inspected under the factor feature x can be obtained

[0101]

[0102] (5-4) Normalize the obtained matching values of the three precision levels according to equation (26), and the reliability distribution e describing the data to be evaluated at different precision levels as shown in equation (25) can be obtained : i

[0103]

[0104] (5-5) Similarly, for different input factor features, the reliability distribution as shown in Table 4 can be obtained;

[0105] Table 4 Reliability distribution under different factor features

[0106] ​

[0107]

[0108] For the sake of easy understanding, the following is an example to illustrate step (5):

[0109] Construct 15 sets of random fuzzy templates The expected curve and the entropy-included expected curve, and match them with the expected curves and entropy-included expected curves of 5 sets of random fuzzy samples to be detected and normalize the obtained matching values into a belief distribution, as shown in Table 5:

[0110] Table 5 Belief distributions under different factor characteristics

[0111]

[0112] The specific description of the above step (6) is as follows:

[0113] (6-1) Based on the historical feature sample S data , construct an initial judgment matrix U:

[0114]

[0115] (6-2) Normalize the initial judgment matrix to obtain the normalized discrimination matrix U′:

[0116]

[0117] In the formula

[0118] (6-3) Calculate the probabilities of the elements in U′ to obtain the probability matrix U″:

[0119]

[0120] In the formula

[0121] (6-4) Calculate the information entropy b of each input feature through formula (30) i , and calculate the fusion factor W=(w 1 , w 2 ,..., w I ) of each attribute through formula (31):

[0122]

[0123] (6-5) Based on the belief obtained in step (5-5) where M={m 1 , m 2, m 3 }, set the credibility factor r of e i to be w i = w i and fuse the evidence obtained under different features using the belief fusion model to obtain the fused belief distribution O(e):

[0124]

[0125]

[0126] where P(M) is the power set of M is the final belief distribution obtained after fusing 5 belief distributions, and B i , C i is any set different from M

[0127] (6 - 6) Based on the fused belief O(e) obtained in step (6 - 5), select the accuracy level corresponding to the maximum belief as the result of the on - line detection accuracy assessment.

[0128] For easy understanding, an example of step (6) is given here, and the process is as follows:

[0129] Based on the analysis of historical samples, the fusion factor W = (w 1 , w 2 , w 3 , w 4 , w 5 ) can be obtained. Set the credibility factor r of e i to be w i = w i , fuse the belief distributions under different factor features to obtain O(e) = [0.6249, 0.2849, 0.0902], and select the accuracy level m 1 corresponding to 0.6249 as the result of the on - line detection accuracy assessment.

[0130] The specific description of step (7) is as follows:

[0131] (7 - 1) According to the on - line detection accuracy assessment result m d in step (6 - 6), this accuracy level represents the measurement error interval that the measuring device may generate, denoted as where and are the upper and lower bounds of the accuracy level respectively.

[0132] (7 - 2) Use the measuring device to measure the product to be measured, and obtain the actual geometric size of the product The expected geometric size measurement value of the product is And based on the measurement error range corresponding to the accuracy level, calculate and evaluate the evaluation size range of the product to be detected

[0133] (7-3) According to the product design requirements, set the ideal tolerance range of qualified products as Among them, is the lower tolerance, represents the upper tolerance, and obtain the actual tolerance range of the product to be detected according to formulas (33) and (34)

[0134]

[0135]

[0136] (7-4) Based on the actual tolerance range of the product to be detected obtained in step (7-3), if both are satisfied, then the product to be detected is a qualified product, otherwise it is an unqualified product.

[0137] For the sake of easy understanding, an example is given here to illustrate step (7), and the process is as follows:

[0138] Given the actual product processing information as shown in Table 6:

[0139] Table 6 Product processing information

[0140]

[0141] The online detection accuracy evaluation result is m 1 , the corresponding measurement error range is [-0.001, +0.001], calculate and evaluate the evaluation size range of the product to be detected as [9.9381 - 0.001, 9.9381 + 0.001], and obtain the actual tolerance range of the product to be detected as [-0.0156, -0.0136] according to formulas (33) and (34). The product to be detected meets the tolerance requirements and is a qualified product.

[0142] The following combines appendix Figure 2 and appendix Figure 3 , and details the embodiments of the method of the present invention:

[0143] The core part of the present invention is as follows: First, determine the multi-factor fusion model for online detection accuracy evaluation, including the input factor characteristics and the output accuracy level. Through the coarse-grained analysis of historical samples, extract the key parameters of the atomization sample set to generate the atomization sample set. Based on this set, calculate its KL divergence from the historical sample set, and further construct a complete random fuzzy sample set. After obtaining the input factor characteristic data of product quality detection online, generate a corresponding random fuzzy set to be detected. By matching the random fuzzy sample set with the random fuzzy set to be detected, obtain the belief distribution describing the non-linear mapping relationship between multi-environment factor characteristics and detection accuracy. Subsequently, fuse the belief distributions of multiple factor characteristics, and finally determine the accuracy level of online product quality detection. According to the accuracy level, evaluate the actual measurement value, and compare the evaluation value with the upper and lower tolerance ranges of the product geometric dimensions to comprehensively judge whether the product meets the quality standard. Figure 2 The structural block diagram of online detection accuracy evaluation is given in

[0144] The following combines the actual evaluation data to introduce each step of the method of the present invention in detail.

[0145] 1. Determine the input factor characteristics and the output accuracy level of the multi-factor fusion model for online detection accuracy evaluation. During the processing of a certain mechanical component, determine the factor characteristic vector X = [x 1 ,..., x i ,...., x I (I = 5) as the input of the multi-factor fusion model for online detection accuracy evaluation, which respectively represent the external environmental temperature x 1 , the external environmental humidity x 2 , the mechanical amplitude of the detection unit x 3 , the mechanical vibration frequency of the detection unit x 4 and the temperature of the detection unit x 5 during the processing; take the accuracy level {m 1 ,..., m d ,..., m D}(D = 3) as the output of the fusion model, which respectively represent high accuracy m 1 , medium accuracy m 2 , low accuracy m 3 . Here, m 1 represents the detection error interval [-0.001, +0.001], m 2 represents the detection error interval [-0.002, +0.002], and m 3 represents the detection error interval [-0.003, +0.003]. In the processing of this mechanical component, J = 180 historical samples are collected, and they form a characteristic sample set denoted as S data = {[X(j), m d]|j=1,2,...,180; d=1,2,3}.

[0146] 2. Generate atomization sample set:

[0147] The historical sample S data ={[X(j),m d ]|j=1,2,...,180;d=1,2,3}, and the accuracy levels are m respectively under different factor characteristics. 1 、m 2 、m 3 Data Collection For example:

[0148]

[0149] Other collections have similar forms.

[0150] According to steps (2-4), we can get Corresponding atomization sample set For example:

[0151]

[0152]

[0153] Other collections have similar forms.

[0154] 3. Construct a random fuzzy sample set:

[0155] Calculate the historical sample screening The KL divergence value of The KL divergence value is compared to judge The atomization state is further obtained The key parameters of the random fuzzy template set As shown in Table 7:

[0156] Table 7 Random fuzzy sample set Key parameters

[0157]

[0158] Based on the key parameters of the random fuzzy sample set in Table 7, 15 random fuzzy sample sets can be obtained: For example:

[0159]

[0160] Other collections have similar forms.

[0161] 4. Construct a random fuzzy set to be tested:

[0162] When the online evaluation data is the process of obtaining the random fuzzy set to be inspected is as follows: The process is as follows:

[0163] Obtain the key parameters of the random fuzzy set to be inspected according to Equations (19) to (21), as shown in Table 8: as shown in Table 8:

[0164] Table 8 Key parameters of the random fuzzy set to be inspected

[0165]

[0166] Based on the key parameters, the random fuzzy set to be inspected can be obtained For example:

[0167]

[0168] Other sets have a similar form.

[0169] 5. Match the random fuzzy template set with the random fuzzy set to be inspected:

[0170] Construct the expected curves and entropy-included expected curves of 15 random fuzzy template sets, match them with the expected curves and entropy-included expected curves of 5 random fuzzy sets to be inspected, and normalize the obtained matching values into a belief distribution, as shown in Table 9. At the same time, the matching graph between the random fuzzy template set and the random fuzzy set to be inspected is given in as shown in Table 9, and the matching graph between the random fuzzy template set and the random fuzzy set to be inspected is given in Figure 3 6. Fuse the belief distributions and determine the accuracy level of the online product quality inspection according to the fusion results:

[0171] Table 9 Belief distributions under different factor characteristics

[0172]

[0173] 6. Fuse the belief distributions and determine the accuracy level of the online product quality inspection according to the fusion results:

[0174] Based on the analysis of historical samples, the fusion factor W=(w 1 , w 2 , w 3 , w 4 , w 5 ) corresponding to each factor characteristic can be obtained. Set the credibility factor r i of e i =w i . Fuse the belief distributions under different factor characteristics to obtain O(e)=[0.6249, 0.2849, 0.0902], and select the accuracy level m 1As the result of the online detection accuracy evaluation.

[0175] The accuracy level is used to evaluate the actual measured value. By comparing the evaluation value with the upper tolerance and lower tolerance of the geometric dimensions measured for the product, it is determined whether the product is qualified.

[0176] The given actual product processing information is shown in Table 10:

[0177] Table 10 Product Processing Information

[0178]

[0179] The result of the online detection accuracy evaluation is m 1 , and the corresponding measurement error range is [-0.001, +0.001]. Calculate the evaluation dimension range of the product to be detected as [9.9381 - 0.001, 9.9381 + 0.001]. According to Equations (33) and (34), obtain the actual tolerance range of the product to be detected as [-0.0156, -0.0136]. Since the tolerance requirements are met, the product to be detected is a qualified product.

Claims

1. The online detection accuracy assessment method for machining product quality based on multi-factor fusion is characterized by: The following steps are involved: Step 1: Determine the input factor characteristics and output accuracy level for online detection accuracy assessment; The input factor characteristics and output accuracy level are specifically as follows: according to the geometric size type and processing technology of the machined product to be tested, determine the factor characteristic vector X = [x1,...,x i ,....,x I ], I = 5, as the input of online detection accuracy assessment, represents the external environment temperature x1, external environment humidity x2, detection unit mechanical amplitude x3, detection unit mechanical vibration frequency x4 and detection unit temperature x5 during the processing; the accuracy level {m1, ..., m d ,...,m D }, D = 3, as the output, represents high precision m1, medium precision m2, low precision m3; the feature sample set consisting of J historical samples is recorded as S data ={[X(j),m d ]|j=1,2,...,J;d=1,2,...,D}, X(j) is the set of eigenvectors of the jth factor; Step 2: Analyze historical samples, obtain key parameters, and generate a set of atomization samples. The specific implementation process is as follows: Step 2-1: Filter historical samples S according to the same accuracy level corresponding to X data , get the factor feature x i The accuracy level is m d The set of N samples Step 2-2, based on Find about the set Atomization sample collection The key parameters include expected value Ex, entropy En and super entropy He, where super entropy He is one third of entropy En; Step 2-3: Generate a normal random number y with En as the expectation and He as the standard deviation p ; Let Ex be the expectation, |y p | is the standard deviation, generating normal random numbers The corresponding degree of certainty is Get a subset of the atomization template set Steps 2-4 and 2-3 are repeated P times until P normal random numbers and their corresponding degrees of certainty are generated to form a fog sample set. Step 3: Calculate the KL divergence between the atomized sample set and the historical samples, and construct a random fuzzy sample set. The specific implementation process is as follows: Step 3-1: Arrange the P normal random numbers contained in the atomization sample set in ascending order of value to obtain a sorted set Step 3-2: sort the normal random number set Evenly partition into subsets and subset in Step 3-3, similarly, The N sample sets in the and subset in, If N is an odd number, assign the middle element to any of the groups; Step 3-4: Based on subset and subset Calculate KL divergence value Subset-based and subset Calculate KL divergence value judge The degree of atomization Step 3-5: The N samples in are randomly grouped with replacement to obtain S groups of data with n samples in each group. s , where the sth group of data is represented by T s , where S = 1 + log2(P) and n s =P / S; Step 3-6: For the sth group of data T s , calculate its expected value Ex s and standard deviation σ s , and similarly obtain the mean of all S groups of data (Ex1, Ex2, ..., Ex s ,...,Ex S ) and standard deviation (σ1,σ2,...,σ s ,...,σ S ); Step 3-7, obtaining key parameters of the random fuzzy sample set according to the degree of atomization; Step 3-8: Based on the factor feature x obtained in step 3-7 i The accuracy level is m d A random fuzzy template set Corresponding key parameters Then, for the three accuracy levels under the five factor characteristics, the corresponding 15 key parameters of the random fuzzy sample set are obtained; Step 3-9: Based on the key parameters of the random fuzzy sample set, replace Ex, En and He in step 2-3 in turn. According to steps 2-3 and 2-4, 15 random fuzzy sample sets corresponding to the three accuracy levels under the five factor characteristics are obtained. Step 4: Obtain the input factor features of product quality inspection online and construct a random fuzzy set to be inspected; Step 5: Match the random fuzzy sample set with the random fuzzy test set to obtain a reliability distribution describing the nonlinear mapping relationship between the characteristics of multiple environmental factors and the detection accuracy; Step 6: Fuse the reliability distribution and determine the accuracy level of online product quality detection based on the fusion result; Step 7: Use the accuracy level to evaluate the actual measured value and determine whether the product is qualified.

2. The method for online detection accuracy assessment of machining product quality based on multi-factor fusion according to claim 1 is characterized in that: The specific implementation process of steps 3-7 is as follows: like It is judged as non-atomized state, and the key parameters of the random fuzzy sample set are obtained: where q1 = max{Ex1 + 3σ1,..., Ex S + 3σ S}, q2 = max{Ex1 - 3σ1,..., Ex S - 3σ S}, q3 = min{Ex1 + 3σ1,..., Ex S + 3σ S}, q4 = min{Ex1 - 3σ1,..., Ex S - 3σ S}; like It is judged as a fogged state, and the key parameters of the random fuzzy sample set are obtained: Where q5 represents The sample variance of The sample fourth-order central moment of .

3. The method for online detection accuracy assessment of machining product quality based on multi-factor fusion according to claim 2 is characterized in that: The specific implementation process of step 4 is as follows: Step 4-1: Obtain the data to be assessed online Compute random fuzzy test sets Key parameters And determine En based on the key parameters of the random fuzzy template set i and He i ,En i and He i are the means of three corresponding key parameters respectively; Step 4-2: Based on key parameters (Ex i ,,En i ,He i ), replace Ex, En and He in step 2-3, and according to steps 2-3 and 2-4, obtain the random fuzzy test set under the characteristics of 5 factors 4. The method for online detection accuracy assessment of machining product quality based on multi-factor fusion according to claim 3 is characterized in that: The specific implementation process of step 5 is as follows: Step 5-1: Construct a random fuzzy sample set The expected curve and the entropy expectation curve Random fuzzy test set The expected curve V i and the entropy expectation curve Step 5-2: In the factor feature x i Under this condition, we can obtain a set of random fuzzy templates with three levels of accuracy. and random fuzzy test set The boundary value of the intersection area is calculated as follows: calculate and The horizontal coordinate value of the intersection point And the curve and V i The horizontal coordinate value of the intersection point At the same time, obtain the curve and V i The ordinate value of the intersection point And the curve and The vertical coordinate value of the intersection point By obtaining the above four points, the random fuzzy template set is determined and random fuzzy test set The subset range of the intersection area; Step 5-3: According to the range determined in step 5-2, select the random fuzzy template set Select H subset elements satisfy: From the random fuzzy test set Select h subset elements c from i (p), satisfying: In the factor feature x i Down, and The matching value of Step 5-4: Obtain the matching values ​​of the three accuracy levels Normalize to get The three normalized matching values ​​are combined to obtain the reliability distribution of the data to be evaluated at different accuracy levels. i ; Step 5-5: For different input factor characteristics, obtain the corresponding reliability distribution.

5. The method for online detection accuracy assessment of machining product quality based on multi-factor fusion according to claim 4 is characterized in that: The specific implementation process of step 6 is as follows: Step 6-1, the factors of each sample in the historical samples are taken as a row in the matrix, the initial judgment matrix U is constructed, and normalized to obtain the normalized judgment matrix U'; Step 6-2: Calculate the probability of each element in U' to obtain the probability matrix U' and calculate the information entropy b of the input feature from the elements in the probability matrix i , and obtain the fusion factor W of each attribute = (w1,w2,...,w I ): Step 6-3: Based on the reliability obtained in step 5-5 Where M = {m1, m2, m3}, set e i The credibility factor r i =w i , and get the fused belief distribution O(e): Where P(M) is the power set of M, is the confidence distribution after fusion, B i ,C i is any set different from M; The accuracy level corresponding to the maximum fusion confidence is selected as the online detection accuracy assessment result.

6. The method for online detection accuracy assessment of machining product quality based on multi-factor fusion according to claim 5 is characterized in that: The specific implementation process of step 7 is as follows: Step 7-1: According to the online detection accuracy evaluation result m d , the accuracy level represents the measurement error interval produced by the measuring equipment, denoted as in and are the upper and lower bounds of the accuracy level respectively; Step 7-2: Measure the product to be tested and obtain the actual geometric dimensions of the product The expected geometric measurements of the product are And based on the measurement error interval corresponding to the accuracy level, calculate the evaluation size interval of the product to be tested Step 7-3: Set the ideal tolerance range of qualified products as is the lower tolerance, Indicates the upper tolerance and obtains the actual tolerance range of the product to be tested Step 7-4: Based on the actual tolerance range of the product to be tested, if both If the product is found to be qualified, it is a qualified product; otherwise, it is an unqualified product.

Images