Multivariable Bayesian control chart design method with adaptive sampling interval

By introducing adaptive sampling interval and policy iteration algorithms into the multivariable Bayesian control chart, the problem of high sampling costs under equal interval sampling is solved, and better control strategies and higher cost-effectiveness are achieved.

CN118917185APending Publication Date: 2024-11-08HEFEI UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410962866.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing multivariable Bayesian control charts use equally spaced sampling, making it difficult to generate optimal control strategies under high sampling costs.

Method used

Using the multivariable Bayesian control chart design method with adaptive sampling interval, the mathematical model of control strategy optimization problems is constructed under the framework of partially observable semi-Markov decision-making process, and the strategy iterative algorithm is used to obtain the long-term expected average cost. The optimal control strategy of the sampling time is used as the horizontal coordinate, and the posterior probability of the production process being out of control at each sampling time is divided into several intervals to determine whether the shutdown check and the next sampling interval are checked.

Benefits of technology

The probability of out-of-control evaluation of the production process at each sampling moment is realized, and the next sampling plan is adaptively determined, which reduces the sampling cost and improves the cost-effectiveness and calculation speed of the control strategy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917185A_ABST
    Figure CN118917185A_ABST
Patent Text Reader

Abstract

The invention provides a multivariable Bayesian control chart design method and system with an adaptive sampling interval, a storage medium and electronic equipment, and relates to the field of control chart design. The method comprises the following steps: firstly, respectively describing a production process, an observation process and quality control; secondly, under a partially observable semi-Markov decision process framework, constructing a mathematical model of a control strategy optimization problem based on a discretization method; solving the mathematical model by adopting a strategy iterative algorithm again, and obtaining an optimal control strategy for minimizing the long-term expected average cost; and then taking the sampling time as an abscissa, taking the posterior probability that the production process is out of control at each sampling time as an ordinate, dividing a posterior probability interval into a plurality of intervals based on the optimal control strategy, and designing and obtaining the multivariable Bayesian control chart. The control chart can assess the probability of runaway of the production process at each sampling time, thereby adaptively determining the next sampling plan.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of control chart design, and in particular to a multivariate Bayesian control chart design method, system, storage medium and electronic equipment with adaptive sampling interval. Background Art

[0002] Statistical process control plays an important role in improving the quality of production and service processes. As a key tool for statistical process control, control charts are mainly used to monitor the characteristic data of the process and identify changes in characteristic parameters. Specifically, quality-related data is continuously collected through sampling, statistics are calculated and plotted on control charts. Once a certain statistic is found to exceed the control limit, the process is considered out of control, and the process is stopped to eliminate the cause of the process out of control.

[0003] Modern production processes often involve multiple related characteristic parameters. For example, the quality of ultrasonic welding in the production process of new energy batteries can be reflected by parameters such as welding pressure, welding energy, welding amplitude, and welding area. At present, control charts for multivariate detection variables include Hotelling T 2 Control charts, multivariate EWMA charts, multivariate Bayesian control charts, etc. Among them, the multivariate Bayesian control chart (or multivariate Bayesian control chart) has a high sensitivity for detecting the shift of the variable mean vector, which has obvious advantages over other non-Bayesian control charts.

[0004] However, existing multivariate Bayesian control charts mainly use equal-period sampling (or equal-interval sampling), which is difficult to produce optimal control strategies in situations with high sampling costs. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the shortcomings of the prior art, the present invention provides a multivariate Bayesian control chart design method, system, storage medium and electronic device with adaptive sampling intervals, which solves the technical problem that it is difficult to generate an optimal control strategy when using equal-interval sampling for high sampling costs.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A multivariate Bayesian control chart design method with adaptive sampling intervals, comprising:

[0010] The production process, observation process and quality control are described respectively; samples collected at variable sampling intervals are considered;

[0011] In the framework of partially observable semi-Markov decision process, combined with the description of the production process, observation process and quality control, a mathematical model of the control strategy optimization problem is constructed based on the discretization method;

[0012] A policy iteration algorithm is used to solve the mathematical model to obtain an optimal control strategy that minimizes the long-term expected average cost;

[0013] The sampling time is used as the horizontal axis, the posterior probability of the production process being out of control at each sampling time is used as the vertical axis, and the posterior probability interval [0, 1] is divided into several intervals based on the optimal control strategy, and the multivariate Bayesian control chart is designed to obtain the multivariate Bayesian control chart; wherein each interval represents the strategy of determining whether to stop for inspection, and if not, giving the next sampling interval.

[0014] Preferably, the production process, observation process and quality control are described respectively; including:

[0015] (1) Consider a production process X = {X(t), t ≥ 0}, which contains two hidden states: controlled state i = 0 and out-of-control state i = 1. When an abnormality occurs, the production process shifts from state 0 to state 1. The time of abnormality occurrence follows an exponential distribution with a mean of 1 / θ.

[0016] (2) The production process has q quality characteristics, and each sampling can obtain a q-dimensional observation vector; when the system is in state i∈{0,1}, the observation vector y obeys the multivariate normal distribution N q (μ i ,∑), its probability density function is:

[0017]

[0018] Among them, μ0, μ1∈R q Represent the mean vector of the production process in controlled and out-of-control states, ∑∈R q×q represents the covariance matrix; the Mahalanobis distance d between the two mean vectors μ0 and μ1 is

[0019] d=[(μ1-μ0)′∑ -1 (μ1-μ0)] 1 / 2 >0 (2)

[0020] (3) The sampling cost is C S ≥0; in the out-of-control state, due to the increase in the proportion of defective products, the production cost per unit time is M>0; the cost of stopping production and finding abnormalities is C I > 0; if an anomaly is found, the cost of eliminating the anomaly is C R >0, the production process returns to a controlled state after the anomaly is eliminated; the time to find the anomaly is T I≥0, the time to eliminate the abnormality is T R ≥0;

[0021] (4) Considering a variable sampling interval, assume that samples are collected at time S1, S2, ..., where h m =S m -S m-1 represents the sampling interval, and the initial sampling time S0 = 0; at each time point S m The observed data obtained is recorded as Y m ;

[0022] (5) At each sampling time S m , two questions need to be decided in sequence: ① Whether to shut down immediately; ② If not, how long will the next sampling interval be.

[0023] Preferably, the process of constructing the mathematical model of the control strategy optimization problem includes:

[0024] (1) Evaluate the status of the process at the sampling point:

[0025] make

[0026] Π m =Pr(X(S m )=1|S0,S1,Y1,…,S m , Y m ) (3)

[0027] Indicates that at the sampling time S m The posterior probability that the production process is under control and Π0 = 0, then Π m It is used to represent the state of the production process; where Pr(·) represents the probability of an event occurring;

[0028] The posterior probability is updated through sample information at each sampling moment; according to Bayes' theorem, Π m The calculation formula is as follows:

[0029]

[0030] in

[0031] ρ(π,h)=(1-π)exp(-θh) (5)

[0032] ρ(π, h) represents the prior probability that the production process is out of control at the next sampling moment when the probability that the production process is out of control at the current sampling moment is π and the next sampling interval is h.

[0033] Z m =(Y m -μ0)′∑- 1(μ0-μ1) (6)

[0034] Zm represents a univariate variable, which is m )=0, it follows the normal distribution N(0, d 2 ), in X(S m )=1, it follows the normal distribution N(-d 2 , d 2 );

[0035] (2) Constructing a mathematical model for the optimization problem under the framework of partially observable semi-Markov decision process:

[0036] Discretize the posterior probability interval [0, 1] into K equidistant subintervals, and the state space is S = {0, 1, ..., K}; if the posterior probability Π m =0, then Π m In state 0; if Π m ∈(k-1 / K, k / K], then Π m In state k; let

[0037]

[0038] As the representative value of the k∈S state;

[0039] Let action a = 0 to indicate that the process is stopped immediately. It is known that the shortest sampling interval allowed by the production process is δ, and the longest sampling interval is Lδ. Let a = l to indicate that the next sampling interval is lδ, then the action set is defined as A = {0, 1, ..., L}.

[0040] We use ψ:S→A to represent a control strategy that assigns a fixed action a to each state k∈S k ∈A; let Ψ represent the set of all possible strategies, ψ∈Ψ; let λ(ψ) represent the long-term expected average cost under strategy ψ, then the optimal strategy ψ * The conditions met are

[0041]

[0042] Let p k,k′ (a) represents the probability that the process will be in the k′∈S state at the next decision moment if action a∈A is taken in the current state k∈S. Then

[0043]

[0044] Among them, G(z|π,h) means that at the current probability π, after h time units, the posterior probability does not exceed the probability z, which is calculated by the following formula

[0045]

[0046] Among them, Φ(·) represents the cumulative distribution probability of the standard normal distribution;

[0047] Let τ k (a) represents the expected stay time after taking action a∈A in the current state k∈S until the next decision moment, then

[0048]

[0049] Let c k (a) represents the expected cost from taking action a∈A in the current state k∈S until the next decision moment, then

[0050]

[0051] Preferably, the method of solving the mathematical model by using a policy iteration algorithm based on a discretization method to obtain an optimal control strategy that minimizes the long-term expected average cost includes:

[0052] Initialization step: select an arbitrary strategy ψ;

[0053] Strategy evaluation step: Calculate the long-term expected average cost λ(ψ) of strategy ψ and the value function V(k) of each state by solving the following equation:

[0054]

[0055] Where V(0)=0;

[0056] Strategy improvement step: Construct a new strategy ψ′, where each state k∈S corresponds to an action a′ k Satisfy the following formula:

[0057]

[0058] Among them, min is the minimization function;

[0059] Convergence test step: If ψ′=ψ, the algorithm stops and outputs the optimal strategy ψ * =ψ; otherwise, let ψ = ψ′ and return to the strategy evaluation step to continue optimization.

[0060] Preferably, in the multivariate Bayesian control chart, changes in grayscale are used to distinguish different intervals.

[0061] A multivariate Bayesian control chart design system with adaptive sampling intervals, comprising:

[0062] A description module is used to describe the production process, observation process and quality control respectively, wherein the collection of samples at variable sampling intervals is considered;

[0063] A modeling module is used to construct a mathematical model of the control strategy optimization problem based on a discretization method in the framework of a partially observable semi-Markov decision process, combined with descriptions of the production process, the observation process and the quality control;

[0064] An optimization module, used for solving the mathematical model by using a policy iteration algorithm to obtain an optimal control strategy that minimizes the long-term expected average cost;

[0065] A design module is used to take the sampling time as the horizontal coordinate and the posterior probability of the production process being out of control at each sampling time as the vertical coordinate, and divide the posterior probability interval [0, 1] into several intervals based on the optimal control strategy, so as to design and obtain the multivariate Bayesian control chart; wherein the strategy represented by each interval is: determining whether to stop for inspection, and if not stopping for inspection, giving the next sampling interval.

[0066] A storage medium stores a computer program for multivariate Bayesian control chart design with adaptive sampling intervals, wherein the computer program enables a computer to execute the multivariate Bayesian control chart design method as described above.

[0067] An electronic device, comprising:

[0068] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including methods for executing the multivariate Bayesian control chart design method as described above.

[0069] (III) Beneficial effects

[0070] The present invention provides a multivariate Bayesian control chart design method, system, storage medium and electronic device with adaptive sampling interval. Compared with the prior art, it has the following beneficial effects:

[0071] In the present invention, the production process, the observation process and the quality control are described respectively first; secondly, under the framework of the partially observable semi-Markov decision process, a mathematical model of the control strategy optimization problem is constructed based on the discretization method to ensure the discreteness and finiteness of the system state space and action set; the policy iteration algorithm is used to solve the mathematical model again to obtain the optimal control strategy that minimizes the long-term expected average cost; then the sampling time is used as the horizontal coordinate, the posterior probability of the production process out of control at each sampling time is used as the vertical coordinate, and the posterior probability interval is divided into several intervals based on the optimal control strategy, and the multivariate Bayesian control chart is designed to obtain. Compared with the traditional multivariate Bayesian control chart, this control chart can evaluate the probability of out of control of the production process at each sampling time, thereby adaptively determining the next sampling plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0073] Figure 1 A block diagram of a multivariate Bayesian control chart design method with adaptive sampling intervals provided by an embodiment of the present invention;

[0074] Figure 2 An example diagram of a multivariate Bayesian control chart with adaptive sampling intervals provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] The embodiments of the present application solve the technical problem that it is difficult to generate an optimal control strategy for high sampling costs when using equal-interval sampling by providing a multivariate Bayesian control chart design method, system, storage medium and electronic device with adaptive sampling intervals.

[0077] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0078] The multivariate Bayesian control chart design method with adaptive sampling interval provided by the present invention at least involves the following core points:

[0079] (1) The designed multivariate Bayesian control chart with adaptive sampling interval will make a decision at each sampling moment whether the process needs to be stopped to check for anomalies and determine when to collect the next sample.

[0080] (2) The optimization problem is formulated in the framework of partially observable semi-Markov decision process, and a discretization method is introduced to ensure the discreteness and finiteness of the system state space and action set. On this basis, a corresponding semi-Markov decision process quantity is assigned to each system state, including expected cost, expected stay time and transition probability.

[0081] (3) Compared with the traditional multivariate Bayesian control chart with a single sampling interval, the proposed adaptive Bayesian control chart is more cost-effective and also outperforms those Bayesian control charts with two sampling intervals in terms of computational speed.

[0082] (4) A policy iteration algorithm was developed to determine the optimal strategy that minimizes the long-term expected average cost.

[0083] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0084] Embodiment 1:

[0085] like Figure 1 As shown, an embodiment of the present invention provides a multivariate Bayesian control chart design method with adaptive sampling intervals, comprising:

[0086] S1. Describe the production process, observation process and quality control respectively, taking into account the collection of samples at variable sampling intervals;

[0087] S2. In the framework of partially observable semi-Markov decision process, combined with the description of the production process, observation process and quality control, a mathematical model of the control strategy optimization problem is constructed based on the discretization method;

[0088] S3, using a policy iteration algorithm to solve the mathematical model and obtain an optimal control strategy that minimizes the long-term expected average cost;

[0089] S4. Taking the sampling time as the horizontal axis and the posterior probability of the production process being out of control at each sampling time as the vertical axis, and dividing the posterior probability interval [0, 1] into several intervals based on the optimal control strategy, the multivariate Bayesian control chart is designed to be obtained; wherein the strategy represented by each interval is: determining whether to stop for inspection, and if not stopping for inspection, giving the next sampling interval.

[0090] Compared with the traditional multivariate Bayesian control chart, the quality control chart designed in the embodiment of the present invention can evaluate the out-of-control probability of the production process at each sampling moment, thereby adaptively determining the next sampling plan.

[0091] Next, we will introduce the steps of the above scheme in detail:

[0092] First, the production process, observation process and control problems faced by the embodiments of the present invention are introduced:

[0093] Specifically referring to step S1, the production process, observation process and quality control are described respectively; wherein collecting samples at variable sampling intervals is considered; including:

[0094] (1) Consider a production process X = {X(t), t ≥ 0}, which contains two hidden states: controlled state i = 0 and out-of-control state i = 1. When an abnormality occurs, the production process shifts from state 0 to state 1. The time of abnormality occurrence follows an exponential distribution with a mean of 1 / θ.

[0095] (2) The production process has q quality characteristics, and each sampling can obtain a q-dimensional observation vector; when the system is in state i∈(0, 1}, the observation vector y obeys the multivariate normal distribution N q (μ i ,∑), its probability density function is:

[0096]

[0097] Among them, μ0, μ1∈R q Represent the mean vector of the production process in controlled and out-of-control states, ∑∈R q×q represents the covariance matrix; the Mahalanobis distance d between the two mean vectors μ0 and μ1 is

[0098] d=[(μ1-μ0)′∑ -1 (μ1-μ0)] 1 / 2 >0 (2)

[0099] (3) The sampling cost is C S ≥0; in the out-of-control state, due to the increase in the proportion of defective products, the production cost per unit time is M>0; the cost of stopping production and finding abnormalities is C I > 0; if an anomaly is found, the cost of eliminating the anomaly is C R >0, the production process returns to a controlled state after the anomaly is eliminated; the time to find the anomaly is T I ≥0, the time to eliminate the abnormality is T R ≥0;

[0100] (4) Considering a variable sampling interval, assume that samples are collected at time S1, S2, ..., where h m =S m -S m-1 represents the sampling interval, and the initial sampling time S0 = 0; at each time point S m The observed data obtained is denoted as T m ;

[0101] (5) At each sampling time S m , two questions need to be decided in sequence: ① Whether to shut down immediately; ② If not, how long will the next sampling interval be.

[0102] Secondly, the mathematical model of the control strategy optimization problem is introduced. The process is divided into three parts: first, how to evaluate the state of the process at the sampling point is introduced; second, the optimization algorithm model is constructed; finally, the optimization algorithm is given.

[0103] The first two parts refer to step S2 for details. In the framework of partially observable semi-Markov decision process, combined with the description of the production process, observation process and quality control, a mathematical model of the control strategy optimization problem is constructed based on the discretization method.

[0104] (1) Evaluate the status of the process at the sampling point:

[0105] make

[0106] Π m =Pr(X(S m )=1|S0,S1,Y1,…,S m , Y m ) (3)

[0107] Indicates that at the sampling time S m The posterior probability that the production process is under control and Π0 = 0, then Π m It is used to represent the state of the production process; where Pr(·) represents the probability of an event occurring;

[0108] The posterior probability is updated through sample information at each sampling moment; according to Bayes' theorem, Π m The calculation formula is as follows:

[0109]

[0110] in

[0111] ρ(π,h)=(1-π)exp(-θh) (5)

[0112] ρ(π, h) represents the prior probability that the production process is out of control at the next sampling moment when the probability that the production process is out of control at the current sampling moment is π and the next sampling interval is h.

[0113] Z m =(Y m -μ0)′∑ -1 (μ0-μ1) (6)

[0114] Z m Represents a univariate variable, which is in X(S m )=0, it follows the normal distribution N(0, d 2 ), in X(S m )=1, it follows the normal distribution N(-d 2 , d 2 ).

[0115] (2) Constructing a mathematical model for the optimization problem under the framework of partially observable semi-Markov decision process:

[0116] Discretize the posterior probability interval [0, 1] into K equidistant subintervals, and the state space is S = {0, 1, ..., K}; if the posterior probability Π m =0, then Π m In state 0; if Π m ∈(k-1 / K, k / K], then Π m In state k; let

[0117]

[0118] As the representative value of the k∈S state;

[0119] Let action a = 0 to indicate that the process is stopped immediately. It is known that the shortest sampling interval allowed by the production process is δ, and the longest sampling interval is Lδ. Let a = l to indicate that the next sampling interval is lδ, then the action set is defined as A = {0, 1, ..., L}.

[0120] We use ψ:S→A to represent a control strategy that assigns a fixed action a to each state k∈S k ∈A; let Ψ represent the set of all possible strategies, ψ∈Ψ; let λ(ψ) represent the long-term expected average cost under strategy ψ, then the optimal strategy ψ * The conditions met are

[0121]

[0122] Let p k,k′ (a) represents the probability that the process will be in the k′∈S state at the next decision moment if action a∈A is taken in the current state k∈S. Then

[0123]

[0124] Among them, G(z|π,h) means that at the current probability π, after h time units, the posterior probability does not exceed the probability z, which is calculated by the following formula

[0125]

[0126] Among them, Φ(·) represents the cumulative distribution probability of the standard normal distribution;

[0127] Let τ k (a) represents the expected stay time after taking action a∈A in the current state k∈S until the next decision moment, then

[0128]

[0129] Let c k (a) represents the expected cost from taking action a∈A in the current state k∈S until the next decision moment, then

[0130]

[0131] The third part specifically refers to step S3, using a policy iteration algorithm to solve the mathematical model to obtain the optimal control strategy that minimizes the long-term expected average cost; including:

[0132] STEP 0 (initialization step): select an arbitrary strategy ψ;

[0133] STEP 1 (i.e., strategy evaluation step): Calculate the long-term expected average cost λ(ψ) of strategy ψ and the value function V(k) of each state by solving the following equation:

[0134]

[0135] Where V(0)=0;

[0136] STEP 3 (i.e., strategy improvement step): Construct a new strategy ψ′, where each state k∈S corresponds to an action a′ k Satisfy the following formula:

[0137]

[0138] Among them, min is the minimization function;

[0139] STEP 4 (i.e., convergence test step): If ψ′=ψ, the algorithm stops and outputs the optimal strategy ψ * =ψ; otherwise, let ψ = ψ′ and return to STEP 1 (i.e., the strategy evaluation step) to continue the optimization.

[0140] In order to obtain the optimal control strategy ψ * Afterwards, the embodiment of the present invention then executes:

[0141] In step S4, the sampling time is used as the horizontal axis, the posterior probability of the production process being out of control at each sampling time is used as the vertical axis, and the posterior probability interval [0, 1] is divided into several intervals based on the optimal control strategy, and the multivariate Bayesian control chart is designed to obtain; wherein each interval represents the strategy of determining whether to stop for inspection, and if not, giving the next sampling interval.

[0142] Exemplarily, in this step, the change of grayscale is used to distinguish different intervals in the multivariate Bayesian control chart.

[0143] It can be understood that the use process of this quality control chart is as follows:

[0144] First, the observed data of the production process is obtained, and the posterior probability of the production process being out of control at each sampling moment is calculated; then, the corresponding points corresponding to the sampling moments and the posterior probabilities are plotted on the control chart; finally, the strategy represented by the interval in which the point is located is implemented.

[0145] In order to better understand the advantages of the embodiments of the present invention, the following specific examples are provided:

[0146] A production process includes a controlled state and an uncontrolled state. The time when an abnormality occurs in the controlled state follows an exponential distribution (θ = 0.01). The Mahalanobis distance between μ0 and μ1 The cost parameters are M=5, C S =0.6, C I =50, C R =25. Time parameter T I =2 and T R = 2. The other parameters are set as K = 100 and δ = 1.

[0147] The optimal strategy obtained by running the optimization algorithm is shown in Table 1.

[0148] Table 1 Optimal control strategy

[0149]

[0150] Table 1 shows the state space controlled limits P * =0.67 is divided into two decision areas. If the posterior probability falls into the area (0.67, 1], it is necessary to stop the machine immediately for inspection. For other areas that continue to produce, it is further subdivided into 12 sub-areas. When the posterior probability is 0, the next sampling interval is 13, and the sampling interval will decrease by 1 for every 0.01 increase in the posterior probability until it reaches 0.1. When the posterior probability is in the sub-area (0.10, 0.67], the sampling interval remains at 1. Because, as the posterior probability approaches the control limit, the evidence that the process mean has shifted becomes stronger and stronger, and the next sample should be collected as soon as possible.

[0151] Table 2 gives the observation data and calculation process of the production process under the guidance of the optimal control strategy.

[0152] Table 2 Production and observation process

[0153]

[0154] In Table 2, the first column is the sampling sequence number; the second column is the sampling time, which is determined by the optimal strategy; the third column is the sampling interval; the fourth column is the actual process state; the fifth column is the two-dimensional observation vector; and the sixth column is the calculated posterior probability.

[0155] In order to better demonstrate the application of multivariate Bayesian control with adaptive sampling intervals, the posterior probability in Table 2 and the optimal control strategy in Table 1 are combined to draw Figure 2 . Figure 2 The blue dots in the figure represent the values ​​drawn based on the posterior probability of the true state being 0, while the red squares represent the values ​​drawn based on the posterior probability of the true state being 1. The change in grayscale is used to represent the sampling interval, where the posterior probability shows an upward trend. When it exceeds the control limit of 0.67, it is necessary to stop the machine for inspection. As can be seen in the figure, the initial sampling interval is set to 13. Since the posterior probability of the second sampling is 0.1740, according to the optimal strategy, the next sampling interval is 1. Then the process stops until the 8th sampling interval, and the sampling interval is fixed to 1. Therefore, the control chart can adaptively formulate sampling plans and detect problems in the production process in a timely manner.

[0156] Embodiment 2:

[0157] An embodiment of the present invention provides a multivariate Bayesian control chart design system with adaptive sampling intervals, comprising:

[0158] A description module is used to describe the production process, observation process and quality control respectively, in which samples are collected at variable sampling intervals;

[0159] A modeling module is used to construct a mathematical model of the control strategy optimization problem based on a discretization method in the framework of a partially observable semi-Markov decision process, combined with descriptions of the production process, the observation process and the quality control;

[0160] An optimization module, used for solving the mathematical model by using a policy iteration algorithm to obtain an optimal control strategy that minimizes the long-term expected average cost;

[0161] A design module is used to take the sampling time as the horizontal coordinate and the posterior probability of the production process being out of control at each sampling time as the vertical coordinate, and divide the posterior probability interval [0, 1] into several intervals based on the optimal control strategy, so as to design and obtain the multivariate Bayesian control chart; wherein the strategy represented by each interval is: determining whether to stop for inspection, and if not stopping for inspection, giving the next sampling interval.

[0162] Embodiment 3:

[0163] An embodiment of the present invention provides a storage medium storing a computer program for designing a multivariate Bayesian control chart with an adaptive sampling interval, wherein the computer program enables a computer to execute the multivariate Bayesian control chart design method as described in Example 1.

[0164] Embodiment 4:

[0165] An embodiment of the present invention provides an electronic device, including:

[0166] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include methods for executing the multivariate Bayesian control chart design method as described in Example 1.

[0167] It can be understood that the multivariate Bayesian control chart design system with adaptive sampling intervals, storage medium and electronic device provided in the embodiments of the present invention correspond to the multivariate Bayesian control chart design method with adaptive sampling intervals provided in the embodiments of the present invention, and the explanations, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the multivariate Bayesian control chart design method, which will not be repeated here.

[0168] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0169] 1. The embodiment of the present invention designs a multivariate Bayesian control chart with adaptive sampling intervals. At the non-periodic sampling moment, a decision is made whether the process needs to be stopped to check the attributable cause and determine when to collect the next sample.

[0170] 2. The adaptive Bayesian control chart proposed in the embodiment of the present invention is more cost-effective than the traditional multivariate Bayesian control chart with a single sampling interval, and performs better in terms of calculation speed than the Bayesian control chart with two sampling intervals.

[0171] 3. The embodiment of the present invention formulates the optimization problem in the framework of a partially observable semi-Markov decision process and introduces a discretization method to ensure the discreteness and finiteness of the system state space and action set.

[0172] 4. The present invention develops a policy iteration algorithm to determine the optimal strategy that minimizes the long-term expected average cost.

[0173] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0174] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multivariate Bayesian control chart design method with adaptive sampling intervals, characterized in that: include: Describe the production process, observation process and quality control respectively; in Consider collecting samples at variable sampling intervals; In the framework of partially observable semi-Markov decision process, combined with the description of the production process, observation process and quality control, a mathematical model of the control strategy optimization problem is constructed based on the discretization method; A policy iteration algorithm is used to solve the mathematical model to obtain an optimal control strategy that minimizes the long-term expected average cost; The sampling time is used as the horizontal axis, the posterior probability of the production process being out of control at each sampling time is used as the vertical axis, and the posterior probability interval [0,1] is divided into several intervals based on the optimal control strategy, and the multivariate Bayesian control chart is designed to obtain the strategy; wherein each interval represents: determining whether to stop for inspection, and if not stopping for inspection, giving the next sampling interval.

2. The multivariate Bayesian control chart design method according to claim 1, characterized in that: The production process, observation process and quality control are described respectively; including: (1) Consider a production process X = {X(t), t ≥ 0}, which contains two hidden states: controlled state i = 0 and out-of-control state i = 1. When an abnormality occurs, the production process shifts from state 0 to state 1. The time of abnormality occurrence follows an exponential distribution with a mean of 1 / θ. (2) The production process has q quality characteristics, and each sampling can obtain a q-dimensional observation vector; when the system is in state i∈{0,1}, the observation vector y obeys the multivariate normal distribution N q (μ i ,∑), its probability density function is: Among them, μ0,μ1∈R q Represent the mean vector of the production process in controlled and out-of-control states, Σ∈R q×q represents the covariance matrix; the Mahalanobis distance d between the two mean vectors μ0 and μ1 is d=[(μ1-μ0)′Σ -1 (μ1-μ0)] 1 / 2 >0 (2) (3) The sampling cost is C S ≥0; in the out-of-control state, due to the increase in the proportion of defective products, the production cost per unit time is M>0; the cost of stopping production and finding abnormalities is C I >0; if an anomaly is found, the cost of eliminating the anomaly is C R >0, the production process returns to a controlled state after the anomaly is eliminated; the time to find the anomaly is T I ≥0, the time to eliminate the abnormality is T R ≥0; (4) Considering a variable sampling interval, assume that samples are collected at time S1, S2, …, where h m =S m -S m-1 represents the sampling interval, and the initial sampling time S0 = 0; at each time point S m The observed data obtained is recorded as Y m ; (5) At each sampling time S m , two questions need to be decided in sequence: ① Whether to shut down immediately; ② If not, how long will the next sampling interval be.

3. The multivariate Bayesian control chart design method according to claim 2, characterized in that: The process of constructing the mathematical model of the control strategy optimization problem includes: (1) Evaluate the status of the process at the sampling point: make Π m =Pr(X(S m )=1|S0,S1,Y1,…,S m ,Y m ) (3) Indicates that at the sampling time S m The posterior probability that the production process is under control and Π0 = 0, then Π m It is used to represent the state of the production process; where Pr(·) represents the probability of an event occurring; The posterior probability is updated through sample information at each sampling moment; according to Bayes' theorem, Π m The calculation formula is as follows: in ρ(π,h)=(1-π)exp(-θh) (5) ρ(π,h) represents the prior probability that the production process is out of control at the next sampling moment when the probability that the production process is out of control at the current sampling moment is π and the next sampling interval is h. Z m =(Y m -μ0)′Σ -1 (μ0-μ1) (6) Z m Represents a univariate variable, which in X(S m )=0, it follows the normal distribution N(0,d 2 ), in X(S m )=1, it follows the normal distribution N(-d 2 ,d 2 ); (2) Constructing a mathematical model for the optimization problem under the framework of partially observable semi-Markov decision process: Discretize the posterior probability interval [0,1] into K equally spaced subintervals, and the state space is S = {0,1,...,K}; if the posterior probability Π m =0, then Π m In state 0; if Π m ∈(k-1 / K,k / K], then Π m In state k; let As the representative value of the k∈S state; Let action a = 0 to indicate that the process is stopped immediately. It is known that the shortest sampling interval allowed by the production process is δ, and the longest sampling interval is Lδ. Let a = l to indicate that the next sampling interval is lδ, then the action set is defined as A = {0, 1, ..., L}. We use ψ:S→A to represent a control strategy that assigns a fixed action a to each state k∈S k ∈A; let Ψ represent the set of all possible strategies, ψ∈Ψ; let λ(ψ) represent the long-term expected average cost under strategy ψ, then the optimal strategy ψ * The conditions met are Let p k,k′ (a) represents the probability that the process will be in the k′∈S state at the next decision moment if action a∈A is taken in the current state k∈S. Then Where G(z|π,h) represents the probability that the posterior probability does not exceed z after h time units at the current probability π, which is calculated by the following formula Among them, Φ(·) represents the cumulative distribution probability of the standard normal distribution; Let τ k (a) represents the expected stay time after taking action a∈A in the current state k∈S until the next decision moment, then Let c k (a) represents the expected cost from taking action a∈A in the current state k∈S until the next decision moment, then 4. The multivariate Bayesian control chart design method according to claim 3, characterized in that: The method of using a discretization-based strategy iteration algorithm to solve the mathematical model and obtain an optimal control strategy that minimizes the long-term expected average cost includes: Initialization step: select an arbitrary strategy ψ; Strategy evaluation step: Calculate the long-term expected average cost λ(ψ) of strategy ψ and the value function V(k) of each state by solving the following equation: Where V(0)=0; Strategy improvement step: Construct a new strategy ψ′, where each state k∈S corresponds to an action a′ k Satisfy the following formula: Among them, min is the minimization function; Convergence test step: If ψ′=ψ, the algorithm stops and outputs the optimal strategy ψ * =ψ; otherwise, let ψ = ψ′ and return to the strategy evaluation step to continue optimization.

5. The multivariate Bayesian control chart design method according to claim 4, characterized in that: In the multivariate Bayesian control chart, the change of gray scale is used to distinguish different intervals.

6. A multivariate Bayesian control chart design system with adaptive sampling intervals, characterized in that: include: Description module, used to describe the production process, observation process and quality control respectively; in Consider collecting samples at variable sampling intervals; A modeling module is used to construct a mathematical model of the control strategy optimization problem based on a discretization method in the framework of a partially observable semi-Markov decision process, combined with descriptions of the production process, the observation process and the quality control; An optimization module, used for solving the mathematical model by using a policy iteration algorithm to obtain an optimal control strategy that minimizes the long-term expected average cost; A design module is used to take the sampling time as the horizontal coordinate and the posterior probability of the production process being out of control at each sampling time as the vertical coordinate, and divide the posterior probability interval [0,1] into several intervals based on the optimal control strategy, so as to design and obtain the multivariate Bayesian control chart; wherein the strategy represented by each interval is: determine whether to stop for inspection, and if not, give the next sampling interval.

7. A storage medium, characterized in that: The computer program for designing a multivariate Bayesian control chart with an adaptive sampling interval is stored therein, wherein the computer program enables the computer to execute the multivariate Bayesian control chart design method according to any one of claims 1 to 5.

8. An electronic device, characterized in that: include: one or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including a program for executing the multivariate Bayesian control chart design method as described in any one of claims 1 to 5.