Product quality sampling method based on deep learning

Abstract

The invention discloses a product quality sampling method based on deep learning, and the method comprises the steps: representing the adaptation degree of a worker and a production line, and the adaptation degree of a raw material or an intermediate product and the production line through a target coefficient; the target coefficient is applied to the subsequent sampling step, so that the adaptation degree can play a certain role in guiding the formulation, modification and execution of the sampling strategy, the sampling result can be matched with the order scale and the artificial error through the technical means, and then the spot check can be executed foreseeably in the generation link. Furthermore, the application also adopts an LSTM prediction model, and the LSTM prediction model is obtained based on historical data training of the production line, so that the LSTM prediction model can master the equipment condition of the production line and the adaptation condition of the equipment and the raw materials, and the sampling result can be matched with the equipment factors and the raw material factors through the technical means. Therefore, the obtained sampling strategy is more comprehensive and is more matched with the actual situation.

Description

Technical Field

[0001] This application relates to the field of electro-digital data processing technology, and in particular, to a method for digital calculation or data processing specifically applicable to specific applications, specifically a product quality sampling method based on deep learning. Background Art

[0002] Multiple signs indicate that the order scale of individual distributors is usually not too large, which leads to a relatively serious error when implementing sampling in a relatively small number of products, such as hundreds or thousands, in the related technology where the product quality sampling method completely relies on statistical laws. Compared with large-scale orders, small-scale orders usually have lower profits. If product quality problems cannot be detected in a timely manner during the production process, it may cause relatively large losses. Moreover, for small-scale orders, real economy enterprises usually do not build a separate production line for it, but rent an existing production line and recruit temporary workers. This leads to the adaptability between the production line and the product and the familiarity of workers with the production line will more or less affect the product quality, inevitably there are potential hazards, and it also makes the technical means adopted during product quality sampling need to be more cautious.

[0003] In view of this, how to perform sampling with predictability of product quality problems for such small-scale orders has become an urgent problem to be solved.

[0004] For example, the patent application with the publication (announcement) number: CN115543260A and the patent title: "A method for randomly extracting samples applied to waste paper quality inspection" (main classification number: G06F7 / 58) realizes the informatization and automation of waste paper number extraction through the cooperation of isolation rules and sampling rules, so as to achieve fairness, justice, and openness in waste paper number extraction. On the one hand, it can illustrate that the electro-digital data processing technology has great potential in the related technical field of product sampling; on the other hand, it can also illustrate that there is a relatively broad expansion prospect for technical exploration in this field. Summary of the Invention

[0005] The embodiments of this application provide a product quality sampling method based on deep learning to at least partially solve the above technical problems.

[0006] The embodiments of this application adopt the following technical solutions: In a first aspect, the embodiments of this application provide a product quality sampling method based on deep learning, and the method includes: When it is detected that the target production line has received an order task, for each activated process node of the target production line, a full inspection is performed on a specified number of intermediate products produced by it to obtain the target coefficient corresponding to each process node; the target coefficient is positively correlated with the production efficiency of the corresponding intermediate product and is also positively correlated with the yield rate of the intermediate product. Using the LSTM prediction model pre-trained based on historical data, and the detection standard adopted for the order task, the data of the raw materials used, and the quantity of the target products included in the order task, predict the expected yield rate of the products corresponding to each process node. When the expected yield rate of the product is not less than the preset yield threshold, use the target coefficient to weight the expected yield rate of the product to obtain the target yield rate corresponding to each process node. When the target yield rates corresponding to each process node show a gradually changing trend along the execution order of the processes, use a preset sampling strategy to conduct spot checks on the intermediate products and target products.

[0007] In an optional embodiment of this specification, the method further includes: When the target yield rate of a certain process node shows a sharp downward trend, regard this process node and its next process node as risk process nodes. Increase the sampling rate for the risk process nodes.

[0008] In an optional embodiment of this specification, the method further includes: The target coefficient is also negatively correlated with the quantity of the target products included in the order task and is negatively correlated with the duration of the intermediate warehousing link in the execution plan process of the order task.

[0009] In an optional embodiment of this specification, the method further includes: When the quantity of the target products is less than the preset first quantity threshold and the number of activated process nodes is less than the preset second quantity threshold, the target coefficient is less than the target coefficient when the number of activated process nodes is not less than the second quantity threshold.

[0010] In an optional embodiment of this specification, the method further includes: When the quantity of the target products is not less than the first quantity threshold and the number of activated process nodes is less than the second quantity threshold, the target coefficient is greater than the target coefficient when the number of activated process nodes is not less than the second quantity threshold.

[0011] In an optional embodiment of this specification, the method further includes: The first quantity threshold is positively correlated with the lowest value in the production efficiency.

[0012] In an alternative embodiment of the present specification, the method further includes: If the expected yield rate of the product is less than a preset yield threshold, a prompt message is issued; the prompt message is used to inform of adjusting the production plan of the order task.

[0013] In an alternative embodiment of the present specification, the LSTM prediction model is trained through the following steps: Based on historical order tasks in history, construct its corresponding sample sequence and sample annotation; the sample sequence includes a plurality of sample nodes; the sample nodes correspond one-to-one with the process nodes started by the historical order task; the data included in the sample nodes are: the equipment attribute data, raw material attribute data, applicable detection standards of the process nodes corresponding thereto, and the quantity of products included in the historical order task; the sample annotation represents the product yield rate of the sample nodes. Based on the sample sequence and sample annotation, train a preset LSTM model to obtain the LSTM prediction model.

[0014] In an alternative embodiment of the present specification, the method further includes: The specified quantity is positively correlated with the quantity of the started process nodes and negatively correlated with the distance between the started key process nodes and the last process node in the target production line; the key process nodes are the process nodes with the lowest yield rate determined based on historical data in the target production line.

[0015] In an alternative embodiment of the present specification, the method further includes: The quantity of products included in the historical order task to which the historical data that can be used for sample construction belongs is not less than a preset third quantity threshold; Wherein, the third quantity threshold is negatively correlated with the quantity of process nodes in the production line started by the historical order task; and, the third quantity threshold in the case where the historical order task is paused due to equipment failure is greater than the third quantity threshold in the case where no pause occurs.

[0016] In a second aspect, an embodiment of the present application further provides a product quality sampling device based on deep learning, and the device is used to implement the method in the first aspect.

[0017] In a third aspect, an embodiment of the present application further provides an electronic device, including: A processor; and A memory arranged to store computer-executable instructions, and when the executable instructions are executed, the processor executes the method steps described in the first aspect.

[0018] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium storing one or more programs, which when executed by an electronic device including a plurality of application programs, cause the electronic device to execute the method steps described in the first aspect.

[0019] The above at least one technical solution adopted in the embodiments of the present application can achieve the following beneficial effects: The method in this specification realizes product quality sampling for small-scale orders. During the execution of the method in the present application, the target coefficient characterizes the degree of adaptation between workers and production lines, and the degree of adaptation between raw materials or intermediate products and production lines. The application of the target coefficient to subsequent sampling steps enables this degree of adaptation to play a guiding role in the formulation, modification, and execution of sampling strategies. Therefore, through technical means, the sampling results can be matched with the order scale and human errors, and thus, during the production process, predictive spot checks can be carried out in a timely manner to discover problems and provide conditions for solving problems. Further, the present application also adopts an LSTM prediction model, which is trained based on the historical data of the production line. The LSTM prediction model can master the equipment conditions of the production line and the adaptation conditions between the equipment and raw materials. Therefore, through technical means, the sampling results can be matched with equipment factors and raw material factors, and the resulting sampling strategy is more comprehensive and more in line with the actual situation. The method of the present application realizes the application of digital data processing technology in the product quality sampling method based on deep learning through technical means. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments and descriptions thereof of the present application are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings: Figure 1 is a process schematic diagram of a product quality sampling method based on deep learning provided by an embodiment of this specification; Figure 2 is a schematic structural diagram of an electronic device in an embodiment of this specification. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The present invention will be further described in detail below in conjunction with the specific embodiments and the accompanying drawings. Similar elements in different embodiments are denoted by related similar element numbers. In the following embodiments, many detailed descriptions are provided to enable a better understanding of the present application. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification, which is to avoid the core part of the present application being overwhelmed by excessive descriptions. For those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the descriptions in the specification and the general technical knowledge in the art.

[0022] In addition, the features, operations, or characteristics described in the specification can be combined in any appropriate manner to form various embodiments. At the same time, the steps or actions in the method description can also be reordered or adjusted in a manner obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for clearly describing a certain embodiment and do not mean that they are the necessary sequences, unless it is stated that a certain sequence must be followed.

[0023] The serial numbers assigned to the components in this document, such as "first", "second", etc., are only used to distinguish the described objects and do not have any sequential or technical meaning. And the "connection" and "coupling" mentioned in this application, unless otherwise specified, both include direct and indirect connection (coupling).

[0024] The technical solutions provided by the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings.

[0025] As Figure 1 shown, the product quality sampling method based on deep learning in this specification includes the following steps: S100: When it is detected that the target production line receives an order task, for each activated process node of the target production line, a specified number of intermediate products produced by it are inspected one by one to obtain the target coefficients corresponding to each process node.

[0026] The target production line in this specification is the production line used when executing an order task. Since the scale of the order task is not large, the number of target products it contains may be only thousands or even hundreds, and it is not necessary to build a separate production line for it. The target production line is a production line leased from other companies or a production line used to produce other products that is called.

[0027] The target products included in the order tasks in this specification can be determined according to the actual situation. For example, socks, shoes, underwear, etc. In the case where the target product is underwear, the target production line can include, but is not limited to, one or several of the following equipment: cutting equipment, sewing equipment, cup shaping equipment, sponge processing equipment, punching and trimming equipment, pressing equipment, jacquard equipment, packaging equipment. There is a corresponding relationship between the equipment and the process nodes, and at least one piece of equipment corresponds to a certain process node. Due to the certain flexibility of the order tasks, it is not necessarily guaranteed that all the equipment in the target production line will be enabled during the execution of the order tasks. For example, if the target product is plain underwear without pattern, the jacquard equipment will not be enabled, nor will the process node to which it belongs be started. Unless otherwise specified, the process nodes mentioned below all refer to the started process nodes.

[0028] In an alternative embodiment of this specification, the specified quantity can be an empirical value. Since the target production line has just been started and the workers have just begun to execute production, this is also a period when problems are relatively concentrated. Conducting a full inspection of the intermediate products can discover problems to a large extent. The specified quantity is the quantity for inspecting the output of each process node. For example, if the specified quantity is 10, then a full inspection is carried out on the first 10 intermediate products or target products produced by process node a, process node b, and process node c.

[0029] The target coefficients in this specification correspond one by one to the process nodes. The process nodes correspond one by one to the intermediate products or target products. The target coefficient is positively correlated with the production efficiency of the corresponding intermediate product and also positively correlated with the yield rate of the intermediate product. It is a value between 0 and 1, representing the degree of fit between the workers and the production line, and the degree of fit between the raw materials and the target production line. Production efficiency can reflect the proficiency of workers in operating equipment. Since the order size is small, it is very likely that the workers are temporarily recruited. Also, due to the low profit of small-scale orders, real economy enterprises are unlikely to hire highly skilled workers with high labor costs. Usually, without large-scale and continuous training opportunities, it is difficult for workers to quickly master a certain piece of equipment. For example, jacquard equipment requires a high level of proficiency from workers. That is to say, throughout the entire execution process of the order task, the proficiency of workers in operating equipment may not have a significant and continuous improvement. Therefore, it is necessary to consider the situation of the workers. In addition, the target coefficient also represents the degree of fit between the raw materials or a certain intermediate product and the corresponding equipment. If the degree of fit is low, the product yield rate will naturally be low. For example, if the temperature of the steam iron is too high or the temperature rises too quickly, it will have a negative impact on fabrics made of chemical fiber materials and also cause silk to change color. Laser cutting will also damage materials containing silk or wool. Since the target production line has not continuously produced the target product in history, and real economy enterprises have no production experience identical to the order task in history, if wool is mixed in the fabric and laser cutting is performed, it is very easy to cause defects.

[0030] As for the methods for determining the production efficiency and the yield rate of fully inspected products, the technical means in the related art are applicable to this specification under permitted conditions.

[0031] In an optional embodiment of this specification, the target coefficient is also negatively correlated with the quantity of the target product included in the order task (the fewer the quantity of the target product, the shorter the running-in period between the workers and the target production line, and the more persistent the negative impact of worker unfamiliarity) and negatively correlated with the duration of the intermediate warehousing link in the execution plan process of the order task (the longer the duration of the warehousing link, the more discontinuous the production line operation, which may lead to large errors in the prediction results of the LSTM prediction model).

[0032] In addition, in a further optional embodiment of the present specification, when the quantity of the target product is less than a preset first quantity threshold (which can be an empirical value), and when the quantity of the activated process nodes is less than a preset second quantity threshold (which can be an empirical value), the target coefficient is less than the target coefficient when the quantity of the activated process nodes is not less than the second quantity threshold. Since most of the samples used in the training process of the LSTM prediction model are collected for the entire production line, if the production process of the target product involves fewer devices, on the one hand, it may lead to a large difference between the production situation and the samples used in the training model, resulting in distorted model output; on the other hand, since the running-in period between workers and the target production line is short, it is difficult to stabilize the quality. Therefore, more caution should be exercised.

[0033] When the quantity of the target product is not less than the first quantity threshold, and when the quantity of the activated process nodes is less than the second quantity threshold, the target coefficient is greater than the target coefficient when the quantity of the activated process nodes is not less than the second quantity threshold. When the quantity of the target product is not very small, workers and the target production line can complete a certain degree of running-in. The fewer the process nodes and the relatively simpler the target product, the lower the possibility of generating potential problems. It can be seen that by adjusting the target coefficient based on the quantity of the target product and the quantity of the process nodes, the adaptation effect between labor and the production line can be amplified.

[0034] Optionally, the first quantity threshold is positively correlated with the lowest value in the production efficiency. The lower the production efficiency, the less skilled the workers are, and combined with the small quantity of the target product, it can play a role in amplifying the characteristics.

[0035] In an optional embodiment of the present specification, the specified quantity is positively correlated with the quantity of the activated process nodes (to detect product quality problems caused by the correlation between process nodes), and negatively correlated with the distance between the activated key process nodes and the last process node in the target production line (the distance is the quantity of the intervening process nodes); the key process nodes are determined based on historical data and are the process nodes with the lowest yield rate in the target production line (the low yield rate of the key process nodes may be caused by the accumulation of problems in the previous process nodes. The more backward the key process nodes are, the more likely it is that there are undetected problems in the previous process nodes).

[0036] S102: Use the LSTM prediction model pre-trained based on historical data, the detection standards used in the order task, the data of the raw materials used, and the quantity of the target product included in the order task to predict the expected yield rate of the products corresponding to each process node.

[0037] The LSTM prediction model in this specification is used to predict the yield of intermediate products or target products produced at a process node when the production on the production line reaches a steady state. The LSTM prediction model corresponds to the production line one by one. When there are changes in the production line, the model needs to be retrained.

[0038] What the LSTM prediction model learns are the characteristics shown by the attribute data of each device included in the target production line (for example, sewing device a and sewing device b are normal, and sewing device c is prone to thread entanglement problems); when the production on the production line reaches a steady state (the overall production line reaches the best production state, and the negative impact of manual work can be ignored), the interaction relationship between the characteristics shown by the attribute data of raw materials and the device characteristics (for example, sewing device a is suitable for various specifications of threads, but the diameter of the threads suitable for sewing device b and sewing device c should not exceed XX, otherwise there will be failures, etc.); and the yields under different applicable inspection standards (such as the enterprise standard, national standard, EU standard, etc.) of a certain enterprise.

[0039] Since the data used for training the LSTM prediction model are mostly the situations of the production line itself and the cooperation between the production line and raw materials, its prediction results are difficult to reflect the impact caused by manual work.

[0040] LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) architecture mainly used to process sequential data. Although LSTM itself is an artificially designed neural network model, in its design and training process, it does reflect some principles and ideas related to natural laws.

[0041] 1. Dependence relationship of time series. Embodiment in natural laws: Many phenomena in nature have the dependence of time series. For example, the change of weather, the behavior patterns of organisms, the grammatical structure of languages, etc. are all affected by past states. This causal relationship in time is an important manifestation of natural laws. Embodiment in LSTM: LSTM is specifically used to process sequential data and can capture the dependence relationship between each time step in the sequence. Through the recurrent connection, the current state of LSTM depends not only on the current input but also on the state at the previous moment. This design enables LSTM to understand the time dependence in sequential data like a biological brain, such as understanding the grammatical structure of sentences in natural language processing and predicting future trends based on historical data in time series prediction.

[0042] 2 Memory and Forgetting Mechanisms. Embodiment of Natural Laws: When processing information, the biological brain screens information based on its importance. Important information is memorized, and unimportant information is forgotten. This mechanism helps the brain process and store information efficiently, avoiding interference from irrelevant information. Embodiment in LSTM: The core of LSTM is the cell state, which is similar to the memory unit in the biological brain. LSTM controls the inflow, outflow, and forgetting of information through gating mechanisms (input gate, output gate, and forget gate). The role of the forget gate is similar to the forgetting mechanism of the biological brain. It determines which information is no longer important based on the current input and the previous state, and then removes it from the cell state. The input gate is responsible for writing new important information into the cell state, similar to the memory process of the biological brain for new information.

[0043] 3. Adaptive Learning Ability. Embodiment of Natural Laws: Biological systems have a strong adaptive learning ability and can adjust their behaviors and structures according to environmental changes. This ability is the result of biological evolution and an important manifestation of natural laws. Embodiment in LSTM: During the training process, LSTM adjusts the weights of the network through the backpropagation algorithm to learn the patterns in the data. This learning process is adaptive because the network automatically adjusts the weights according to the characteristics of the training data to better fit the data. For example, when processing different types of sequence data, LSTM automatically adjusts the parameters of the gating mechanism to adapt to the characteristics of the data, thus achieving adaptive learning.

[0044] 4. Information Screening and Integration. Embodiment of Natural Laws: When processing information, biological systems extract useful information and ignore irrelevant information through complex screening and integration mechanisms. This mechanism helps improve the efficiency and accuracy of the system. Embodiment in LSTM: The gating mechanism of LSTM is actually a process of information screening and integration. The input gate determines which information is important based on the current input and the previous state and integrates it into the cell state; the forget gate screens out the no-longer-important information and removes it. This screening and integration process enables LSTM to process sequence data efficiently, avoid interference from irrelevant information, and thus improve the performance of the model.

[0045] 5. Conversion of Energy and Information. Embodiment of Natural Laws: In natural systems, the conversion of energy and information is ubiquitous. For example, organisms process and store information by consuming energy, while the transmission and processing of information also affect the energy consumption of organisms. Embodiment in LSTM: Although LSTM is a computational model, its training process also involves the conversion of energy and information. During the training process, computational resources (which can be analogized to energy) are used to process training data (information), and the weights of the network are adjusted through optimization algorithms to learn the patterns in the data. This process of energy-information conversion enables LSTM to achieve complex information processing by consuming certain resources, just like natural systems.

[0046] 6. Stability and Dynamics of Complex Systems. Embodiment of Natural Laws: Complex systems in nature (such as ecological systems, climate systems, etc.) exhibit both stability and dynamics. These systems can maintain stability within a certain range while also responding to external changes. Embodiment in LSTM: The cell state and gating mechanism of LSTM jointly ensure the stability and dynamics of the network. The cell state, controlled by the gating mechanism, can maintain stability to a certain extent, thus avoiding problems such as vanishing gradients; at the same time, the dynamic adjustment of the gating mechanism enables LSTM to flexibly process different sequence data. This combination of stability and dynamics allows LSTM to maintain stability and adapt to changes when processing sequence data, just like complex systems in nature.

[0047] In summary, although the design and training process of the LSTM model are based on artificially designed algorithms and mathematical principles, it does embody the principles and ideas of natural laws in certain aspects. These principles and ideas not only enable LSTM to efficiently process sequence data but also provide inspiration for understanding the information processing mechanisms of natural systems.

[0048] In the related art, any technical means capable of training an LSTM model is applicable to this specification as long as conditions permit. In an optional embodiment of this specification, the process of model training may be as follows: Based on historical order tasks in the past, construct their corresponding sample sequences and sample annotations; (the historical order tasks, sample sequences, and sample annotations correspond one by one) The sample sequence contains several sample nodes; the sample nodes correspond one by one to the process nodes initiated by the historical order tasks; the data contained in the sample nodes are: the equipment attribute data, raw material attribute data, applicable inspection standards of the process nodes corresponding to them, and the quantity of products included in the historical order tasks; the sample annotations represent the product yield rates of the sample nodes. Feature vectors can be constructed based on this data. Any technical means capable of constructing feature vectors in the related art is applicable to this specification as long as conditions permit. Optionally, the quantity of products included in the historical order tasks to which the historical data used for sample construction belongs is not less than a preset third quantity threshold. The third quantity threshold is negatively correlated with the quantity of process nodes in the production line initiated by the historical order task, so as to conduct a more comprehensive evaluation of the entire production line, which is conducive to reflecting the cooperation situation between process nodes in the features. Moreover, the third quantity threshold in the case where the historical order is paused due to equipment failure is greater than that in the case where no pause occurs, so that the LSTM prediction model can learn the influence of the quantity of products included in the order task on the yield rate. Since the quantity of products included in the samples used in the model training process is not too low, on the one hand, it can objectively reflect the real situation of the equipment, and on the other hand, it can also enable the model to learn the influence of the product quantity on the yield rate. In this way, when using the model offline, since the quantity of target products included in small-scale orders is not large, this can cause a confidence difference in the yield rate output by the model due to the quantity, and further enable the output result to reflect the sensitivity of the production line to the product quantity.

[0049] Train a preset LSTM model based on the sample sequence and sample annotation to obtain the LSTM prediction model.

[0050] S104: In the case where the predicted product yield rate is not less than a preset yield rate threshold, weight the predicted product yield rate with the target coefficient to obtain the target yield rate corresponding to each process node.

[0051] There will be as many predicted product yield rates as there are process nodes. That the predicted product yield rate is not less than the preset yield rate threshold means that the predicted product yield rates of all process nodes corresponding to intermediate products or target products are not less than their respective preset yield rate thresholds. For example, in the case where the yield rate of the previous process node is low, and the yield rate of the subsequent process node is high, it will also result in higher costs.

[0052] The weighting in this specification is one-to-one weighting. For example, the target coefficient of process node A is used to weight the predicted yield rate of the products of process node A. Since the target coefficient can reflect the labor situation, the compatibility of the production line with raw materials; the predicted product yield rate can reflect the situation of the production line itself and the best compatibility of the production line with different raw materials, this combination of weighted forms can make the obtained target yield rate more comprehensive and can more accurately reflect the actual achievable situation.

[0053] In an optional embodiment of this specification, when the predicted yield rate of a certain product is less than a preset yield threshold, a prompt message is sent; the prompt message is used to inform the adjustment of the production plan of the order task. The specific adjustment method can be determined based on manual experience. For example, changing a certain fabric to another fabric, or changing one ironing process to two processes.

[0054] S106: When the target yield rates corresponding to the respective process nodes show a gradually changing trend along the execution order of the processes, a preset sampling strategy is used to conduct spot checks on the intermediate products and target products.

[0055] In the actual production process, it is very difficult to conduct strict spot checks on every process node, and in some processes, spot checks are not even carried out directly. Especially for orders with a small scale, the increased cost brought by strict spot checks will further reduce profits. However, the method in this specification conducts limited full inspections (such as covering dozens of products), which can master the situation of each process node at the initial stage of production. Moreover, full inspection can also reflect the connection effect between each process node.

[0056] Gradually changing means that the slope of the nodes of the yield curve (corresponding one-to-one with the process nodes, with the abscissa being the process nodes and the ordinate being the yield) is not less than a preset slope threshold (an empirical value related to the predicted profit rate of the order task). The target yield rates corresponding to the respective process nodes show a gradually changing trend along the execution order of the processes, indicating that the manual influence is controllable and stable and does not pose a greater threat to the overall production. Otherwise, if the yield rate of an intermediate process node continues to decrease and shows a small slope, it indicates that the risk of this process node is relatively high.

[0057] In the related art, any sampling strategy that can be used for sampling under permitted conditions can be used as the preset sampling strategy in this specification, such as stratified sampling, etc.

[0058] In an alternative embodiment of this specification, when the target yield of a certain process node shows a sharp decline (the slope is less than the slope threshold, that is, the absolute value of the slope has a large value but is negative), this process node and its next process node are regarded as risk process nodes; the sampling rate of the risk process nodes is increased to timely detect the risk of defective intermediate products and improve the risk monitoring effect.

[0059] The method in this specification realizes product quality sampling for small-scale orders. During the execution of the method in this application, the adaptation degree between workers and production lines, and between raw materials or intermediate products and production lines is characterized by a target coefficient. The target coefficient is applied to subsequent sampling steps so that this adaptation degree can play a guiding role in the formulation, modification, and execution of sampling strategies. Then, through technical means, the sampling results can be matched with the order scale and human errors, and thus, during the production process, predictive random inspections can be carried out in a timely manner to detect problems in a timely manner and provide conditions for solving problems. Further, this application also adopts an LSTM prediction model, which is trained based on the historical data of this production line. Then, the LSTM prediction model can master the equipment conditions of this production line and the adaptation conditions between the equipment and raw materials, and thus, through technical means, the sampling results can be matched with equipment factors and raw material factors. The resulting sampling strategy is more comprehensive and more in line with the actual situation.

[0060] Furthermore, this specification also provides a product quality sampling device based on deep learning. The device can execute the method in any of the foregoing embodiments and can obtain the same or similar technical effects, which will not be elaborated here.

[0061] Figure 2 It is a schematic structural diagram of an electronic device according to an embodiment of this application. Please refer to Figure 2 , at the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. Among them, the memory may include internal memory, such as high-speed random access memory (Random-Access Memory, RAM), and may also include non-volatile memory, such as at least one disk memory, etc. Of course, the electronic device may also include other hardware required for other services.

[0062] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 2 only a bidirectional arrow is used in

[0063] Memory, which is used to store programs. Specifically, the program can include program code, and the program code includes computer operation instructions. The memory can include a memory and a non-volatile memory, and provides instructions and data to the processor.

[0064] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a product quality sampling device based on deep learning at the logical level. The processor executes the program stored in the memory and is specifically used to execute any of the foregoing product quality sampling methods based on deep learning.

[0065] As described above in this application Figure 1A product quality sampling method based on deep learning disclosed in the illustrated embodiment can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit in the hardware of the processor or instructions in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method.

[0066] The electronic device can also execute Figure 1 a product quality sampling method based on deep learning in Figure 1 and implement the functions of the illustrated embodiment. The embodiments of the present application will not be elaborated herein.

[0067] The embodiments of the present application also propose a computer-readable storage medium that stores one or more programs. The one or more programs include instructions that, when executed by an electronic device including multiple application programs, execute any of the aforementioned product quality sampling methods based on deep learning.

[0068] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0069] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0070] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0071] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0072] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0073] The memory may include non-permanent memory in the computer-readable medium, random access memory (RAM), and / or non-volatile memory in the form of, for example, read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0074] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0075] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity 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, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0076] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0077] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A product quality sampling method based on deep learning, characterized in that, The method includes: When it is detected that the target production line receives an order task, for each activated process node of the target production line, full inspection is performed on a specified number of intermediate products produced by it to obtain the target coefficient corresponding to each process node; the target coefficient is positively correlated with the production efficiency of its corresponding intermediate product and is also positively correlated with the yield rate of this intermediate product; Using the LSTM prediction model pre-trained based on historical data, and the detection standard adopted by the order task, the data of the raw materials used, and the quantity of the target products included in the order task, predict the expected yield rate of the products corresponding to each process node; When the expected product yield rate is not less than the preset yield rate threshold, use the target coefficient to weight the expected product yield rate to obtain the target yield rate corresponding to each process node; When the target yield rates corresponding to each process node show a gradually changing trend along the execution order of the process, use a preset sampling strategy to conduct sampling inspection on the intermediate products and target products.

2. The method according to claim 1, wherein The method further includes: When the target yield rate of a certain process node shows a sharp downward trend, regard this process node and its next process node as risk process nodes; Increase the sampling rate for the risk process nodes.

3. The method according to claim 1, wherein The method further includes: The target coefficient is also negatively correlated with the quantity of the target products included in the order task and is negatively correlated with the duration of the intermediate warehousing link in the execution plan process of the order task.

4. The method according to claim 1, wherein The method further includes: When the quantity of the target products is less than the preset first quantity threshold and the number of activated process nodes is less than the preset second quantity threshold, the target coefficient is less than the target coefficient when the number of activated process nodes is not less than the second quantity threshold.

5. The method according to claim 4, wherein The method further includes: When the quantity of the target products is not less than the first quantity threshold and the number of activated process nodes is less than the second quantity threshold, the target coefficient is greater than the target coefficient when the number of activated process nodes is not less than the second quantity threshold.

6. The method according to claim 4, wherein The method further includes: The first quantity threshold is positively correlated with the lowest value in the production efficiency.

7. The method according to claim 1, wherein The method further includes: When the expected product yield rate is less than the preset yield rate threshold, a prompt message is sent; the prompt message is used to inform of adjusting the production plan of the order task.

8. The method according to claim 1, wherein The LSTM prediction model is obtained through the following steps of training: According to historical order tasks in the past, construct its corresponding sample sequence and sample annotation; the sample sequence includes several sample nodes; the sample nodes are in one-to-one correspondence with the process nodes activated by the historical order task; the data included in the sample nodes are: the equipment attribute data of the process node corresponding to it, the raw material attribute data, the applicable detection standard, and the quantity of the products included in the historical order task; the sample annotation represents the product yield rate of the sample node; Based on the sample sequence and sample annotation, train a preset LSTM model to obtain the LSTM prediction model.

9. The method according to claim 8, wherein The method further includes: The number of products included in the historical order task to which the historical data that can be used for sample construction belongs is not less than a preset third quantity threshold; wherein, the third quantity threshold is negatively correlated with the number of process nodes in the production line started by the historical order task; and, the third quantity threshold in the case where the historical order task is suspended due to equipment failure is greater than the third quantity threshold in the case where no suspension occurs.

10. The method according to claim 1, wherein The method further includes: The specified quantity is positively correlated with the number of started process nodes and negatively correlated with the distance between the started key process nodes and the last process node in the target production line; the key process node is the process node with the lowest yield rate in the target production line determined based on historical data.

Images