TWI737497B - Quality designing method and electrical device - Google Patents

Abstract

A quality designing method includes: training a machine learning model according to history manufacturing parameters and history quality data; establishing an objective function according to a customer request; performing an optimization algorithm based on the objective function so as to try multiple sets of manufacturing parameters, input each set of manufacturing parameters to the machine learning model to obtain predicted quality data, and search the predicted quality data which optimizes the objective function.

Description

Quality design method and electronic device

This disclosure is about automated quality design methods.

The quality parameter design at this stage mostly relies on human knowledge and experience. When customers put forward quality requirements, they need to rely on human experience to measure those manufacturing parameters that can produce products that meet customer requirements, and then determine whether they can accept orders. However, relying on human knowledge and experience mixed with many subjective judgments, how to propose an objective quality design method is a topic of concern to those skilled in the art.

The embodiment of the present invention proposes a quality design method suitable for a computer device, including: obtaining historical manufacturing parameters and historical quality data, training a machine learning model based on the historical manufacturing parameters and historical quality data; receiving customer requirements, and according to customer requirements Establish an objective function; and execute an optimization algorithm based on the objective function to try multiple sets of manufacturing parameters, input each set of manufacturing parameters into the machine learning model to obtain predicted quality data, and And look for prediction quality data that optimizes the objective function.

In some embodiments, the above-mentioned quality design method is applicable to the rolling system. Historical manufacturing parameters include alloy composition, hot rolling temperature, cold rolling temperature, speed, tension, and quenching and tempering rate. Historical quality parameters include tensile strength, yield strength and elongation.

In some embodiments, the step of performing the optimization algorithm according to the objective function further includes: setting restriction conditions so that the alloy composition in each set of manufacturing parameters must be within an existing range.

In some embodiments, if the customer request is a bilateral request, the objective function is set to be a concave function or a convex function.

In some embodiments, if the customer requirement is a unilateral requirement, the objective function is set as a monotonically increasing function or a monotonically decreasing function.

From another perspective, an embodiment of the present invention provides an electronic device including a memory and a processor. The memory stores multiple instructions, and the processor executes these instructions to complete multiple steps: obtain historical manufacturing parameters and historical quality data, train a machine learning model based on historical manufacturing parameters and historical quality data; receive customer requirements and follow Requires the establishment of an objective function; and executes an optimization algorithm based on the objective function to try multiple sets of manufacturing parameters, input each set of manufacturing parameters to the machine learning model to obtain predicted quality data, and find the predicted quality of the optimized objective function data.

In the above-mentioned quality design method, it is possible to automatically calculate the manufacturing parameters that have lower production costs and meet the quality standards, which can speed up the overall efficiency of subsequent trial production and answering whether the order can be accepted.

In order to make the above-mentioned features and advantages of the present invention more comprehensible, the following specific embodiments are described in detail in conjunction with the accompanying drawings.

110: Rolling system

111: Steelmaking stage

112: Hot rolling stage

113: Cold rolling stage

120: electronic device

121: processor

122: memory

210: Convex function

220: monotonically increasing function

230: Monotonic decreasing function

301~303: steps

[Figure 1] is a schematic diagram showing the rolling system.

[FIG. 2A] to [FIG. 2C] are schematic diagrams illustrating the objective function according to an embodiment.

[Fig. 3] is a flowchart of a quality design method according to an embodiment.

Regarding the “first”, “second”, etc. used in this text, it does not particularly mean the order or sequence, but only to distinguish elements or operations described in the same technical terms.

Fig. 1 is a schematic diagram showing the rolling system. 1, the rolling system 110 includes multiple production stages, such as the steelmaking stage 111, the hot rolling stage 112, and the cold rolling stage 113. Each production stage requires corresponding equipment and processes, and has general knowledge in the field. The reader should understand these stages of production, so I won’t repeat them here. In addition, the relevant parameters of these production stages will be transmitted to the electronic device 120, and the electronic device 120 includes a processor 121 and a memory 122. The electronic device 120 can be a control computer in various forms. The processor 121 can be a central processing unit, a microprocessor, a microcontroller, a digital signal processor, a special application integrated circuit, etc., and the memory 122 can be a volatile memory or Non-volatile memory, which stores multiple commands, The processor 121 executes these instructions to complete a quality design method, which will be described in detail below.

First, obtain historical manufacturing parameters and historical quality data about the rolling system 110. The historical manufacturing parameters are about manufacturing processes and ingredients, and the historical quality data are about product quality. For example, the product produced here is a steel coil, and the historical manufacturing parameters include the alloy composition of the steel coil, hot rolling temperature, cold rolling temperature, speed, tension, and quenching and tempering rate. On the other hand, historical quality data includes tensile strength, yield stress and elongation. However, the present invention is not limited to this. In other embodiments, historical manufacturing parameters and historical quality data may also be related to products other than steel coils. In some embodiments, some pre-processing can also be performed on these historical manufacturing parameters and historical quality data. The pre-processing includes outlier filtering and normalization, but the present invention does not limit the content of these pre-processing.

Next, train a machine learning model based on the aforementioned historical manufacturing parameters and historical quality data. The purpose is to predict quality data based on manufacturing parameters. In other words, historical manufacturing parameters are used as input to the machine learning model during the training phase, while historical quality data It is used as the output of the machine learning model. The machine learning model can be a random forest algorithm, a support vector machine, a neural network, etc. The present invention is not limited to this. In some embodiments, the loss function used in the training phase is the mean square error between the predicted value and the ground truth, but the present invention is not limited thereto. In the testing phase, the manufacturing parameters are input to the trained machine learning model, and the output of the machine learning model is the predicted quality data. This calculation can be expressed as y _i = g ( x _i ), where g () represents the trained machine Learning model, x _i represents the i-th group of manufacturing parameters, y _i represents the predicted quality data corresponding to the i-th group of manufacturing parameters, and i is a positive integer. It is worth noting that boldface is used to represent vectors, and non-boldface is scalar.

Next, the customer request is received. This customer request can be about one or more quality data, and can be a unilateral request or a bilateral request. Specifically, a unilateral requirement means that one or more quality data must be greater than a preset value; or a unilateral requirement can also mean that one or more quality data must be less than a preset value. On the other hand, bilateral requirements mean that one or more quality data must be within a preset range, for example, between 10-20, and cannot be less than 10 nor greater than 20.

An objective function can be established according to customer requirements. When the customer's requirement is a bilateral requirement, the objective function is set as a concave (concave) function or a convex (convex) function. For example, if the preset range requested by the customer is [a, b], where a and b are real numbers, which means that a certain quality data must be within this range, then the objective function can be set to the convex function 210 in FIG. 2A. The convex function 210 is positive in the range from the real number a to the real number b, and is 0 in other ranges. The maximum value of the convex function 210 is to be searched for in the subsequent algorithm. Alternatively, the convex function 210 can be multiplied by -1 (to form a concave function), and the minimum value of the objective function can be searched for in the subsequent algorithm. In other words, the choice of concave function or convex function depends on the maximum or minimum value of the objective function to be searched.

When the customer requirements are unilateral requirements, the objective function can be set as a monotonic increasing function or a monotonic decreasing function. For example, if the customer requires that a certain quality data must be greater than a preset value, a monotonic increasing function 220 as shown in Figure 2B can be used. When the input of the monotonic increasing function 220 is less than or equal to c, the output is 0. When the monotonic increasing function When the input of 220 is greater than c, the output is positive, where c is a real number. In the subsequent algorithm, the maximum value of this monotonically increasing function 220 should be searched. In some embodiments, the aforementioned real number c can be set to a preset value required by the customer, or it can be a preset value plus n times the standard deviation (calculated based on historical quality data), n is a positive integer, and this positive integer n is 3 or other suitable numerical values, for example. From another perspective, this objective function can be expressed as the following mathematical formula (1), where y represents quality data. α is a positive real number, which can be determined through experiments.

On the other hand, if the customer requires a certain quality data to be less than a preset value, a monotonic decreasing function 230 as shown in Figure 2C can be used. When the input of this monotonic decreasing function 230 is less than d, the output is positive, and when the monotonic decreasing function 230 When the input of is greater than or equal to d, the output is 0, where d is a real number. In the subsequent algorithm, the maximum value of this monotonically decreasing function 230 should be searched. In some embodiments, the aforementioned real number d can be set to a preset value required by the customer, or it can be a preset value minus n times the standard deviation (calculated based on historical quality data). The positive integer n is, for example, 3 or Other suitable values. From another perspective, the objective function can be expressed as the following mathematical formula (2), where β is a positive real number, which can be determined through experiments.

[Math 2]

In the embodiment of FIG. 2B and FIG. 2C, it is necessary to search for the maximum value of the objective function. To search for the minimum value, a monotonically decreasing function can be used when the customer requires that a certain quality data must be greater than a preset value. When a certain quality data must be less than a preset value, a monotonically increasing function is used. In other words, the choice of monotonic increasing function or monotonic decreasing function depends on the maximum or minimum value of the objective function to be searched, and also depends on which unilateral requirement the customer request belongs to.

The convex function 210, the monotonically increasing function 220, and the monotonically decreasing function 230 in FIGS. 2A to 2C are just examples. Those skilled in the art should design other functions based on the above disclosure.

After the objective function is determined, an optimization algorithm can be executed according to the objective function to try multiple sets of manufacturing parameters, input each set of manufacturing parameters into the above-mentioned machine learning model to obtain predicted quality data, and find the best Optimize the predicted quality data of the objective function. In some embodiments, restrictions can also be set when the optimization algorithm is executed, so that the alloy composition in each set of manufacturing parameters must be within the existing range. Specifically, this optimization algorithm can be expressed as the following mathematical formula (3).

為一集合，表示上述的既有範圍，此集合

包括所有曾經生產過的合金成份。f()是上述的目標函數。值得注意的是，當 預測品質數據 y _i 為向量(長度大於1)時，可以將每一個預測品質數據都輸入對應的目標函數，然後將結果加總起來。舉例來說，如果客戶要求第一品質數據必須在[a,b]的範圍內，第二品質數據必須大於c，第三品質數據必須小於d時，則所預測出的第一至第三品質數據可以分別輸入至凸函數210、單調遞增函數220與單調遞減函數230，然後再將這些函數的輸出加總起來做為數學式(3)中的f()。 Where ac _i represents the alloy composition in the i-th group of manufacturing parameters x _i.

Is a set, representing the above-mentioned existing range, this set

Including all the alloy components that have been produced. f () is the above objective function. It is worth noting that when the predicted quality data y _i is a vector (the length is greater than 1), each predicted quality data can be input to the corresponding objective function, and then the results can be added up. For example, if the customer requires that the first quality data must be within the range of [a,b], the second quality data must be greater than c, and the third quality data must be less than d, then the predicted first to third quality The data can be input into the convex function 210, the monotonically increasing function 220, and the monotonically decreasing function 230 respectively, and then the outputs of these functions are added together as f () in the mathematical formula (3).

In some embodiments, biological heuristic algorithms, such as genetic algorithms or Particle Swarm Optimization (PSO) can be used to search for manufacturing parameters x _i to find the best set of manufacturing parameters x _i , The corresponding predicted quality data y _i enables the objective function to output the maximum value. Those with ordinary knowledge in the art should understand how to use the biological heuristic algorithm under a given objective function, which will not be described in detail here. In this optimization algorithm, limiting the alloy composition must be within the existing range in order to slow down the increase of rigid species, increase the continuous casting rate and reduce the cost, so that the manufacturing parameters with lower production costs and quality standards can be searched for.

Fig. 3 is a flowchart illustrating a quality design method according to an embodiment. Referring to FIG. 3, in step 301, historical manufacturing parameters and historical quality data are obtained, and a machine learning model is trained based on the historical manufacturing parameters and historical quality data. In step 302, customer requirements are received, and an objective function is established according to the customer requirements. In step 303, an optimization algorithm is executed according to the objective function to try multiple sets of manufacturing parameters, each set of manufacturing parameters is input to the machine learning model to obtain predicted quality data, and the predicted quality data of the optimized objective function is searched. However, each step in Figure 3 has been described in detail as above, here I won't repeat it. It is worth noting that each step in FIG. 3 can be implemented as multiple program codes or circuits, and the present invention is not limited thereto. In addition, the method in FIG. 3 can be used in conjunction with the above embodiments, or can be used alone. In other words, other steps can also be added between the steps in FIG. 3.

In the above-mentioned quality design method, it is possible to automatically calculate the manufacturing parameters with low production cost and meet the quality standards, which can provide product engineers with decision-making assistance in quality design, speed up subsequent trial production and reply to the overall efficiency of whether to accept orders.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention. Anyone with ordinary knowledge in the relevant technical field can make some changes and modifications without departing from the spirit and scope of the present invention. The protection scope of the present invention shall be subject to those defined by the attached patent application scope.

301~303: steps

Claims (4)

A quality design method suitable for an electronic device, including: obtaining historical manufacturing parameters and historical quality data, training a machine learning model based on the historical manufacturing parameters and the historical quality data; receiving a customer request, and building a machine learning model based on the customer request Objective function, if the customer requirement belongs to a two-sided requirement, set the objective function as a concave function or a convex function, if the customer requirement belongs to a unilateral requirement, set the objective function as a monotonically increasing function or a monotonous decreasing function; And execute an optimization algorithm according to the objective function to try multiple sets of manufacturing parameters, input each set of manufacturing parameters to the machine learning model to obtain predicted quality data, and set a restriction condition so that each set The alloy composition in the manufacturing parameters must be within an existing range, and the predicted quality data that optimizes the objective function is sought. The quality design method according to claim 1, wherein the quality design method is suitable for a rolling system, and the historical manufacturing parameters include alloy composition, hot rolling temperature, cold rolling temperature, speed, tension, and quenching and tempering rate, and the historical quality Parameters include tensile strength, yield strength and elongation. An electronic device includes: a memory storing a plurality of instructions; and a processor for executing the instructions to complete a plurality of steps: obtaining historical manufacturing parameters and historical quality data, and according to the historical manufacturing parameters Training a machine learning model with the historical quality data; receive a customer request, and establish an objective function according to the customer’s request. If the customer’s request belongs to a two-sided requirement, set the objective function to be a concave function or a convex function. The customer requirement is a unilateral requirement. Set the objective function as a monotonically increasing function or a monotonically decreasing function; and execute an optimization algorithm according to the objective function to try multiple sets of manufacturing parameters. Input manufacturing parameters to the machine learning model to obtain predicted quality data, set a restriction condition so that the alloy composition in each set of manufacturing parameters must be within an existing range, and search for the predicted quality that optimizes the objective function data. The electronic device according to claim 3, wherein the quality design method is suitable for a rolling system, and the historical manufacturing parameters include alloy composition, hot rolling temperature, cold rolling temperature, speed, tension, and quenching and tempering rate, and the historical quality parameters Including tensile strength, yield strength and elongation.

Images