Data analysis device and method
The data analysis device uses a line takt estimation model to analyze assembly line operations, addressing the challenge of efficiency fluctuations by providing accurate identification of performance declines and their causes.
Patent Information
- Application Number
- JP2022527483
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-05-27
- Filing Date
- 2020-11-24
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2040-11-24
AI Technical Summary
Existing technologies struggle to easily analyze fluctuations in the efficiency of assembly line operations across multiple processes, making it difficult to identify the cause of performance declines in production lines.
A data analysis device and method that utilizes a line takt estimation model based on Bayesian estimation to analyze equipment log data, generating probability distributions for process efficiency and factor variations, allowing for the detection and evaluation of malfunctions and their causes.
Enables easy analysis of efficiency fluctuations and identification of factors contributing to performance declines in assembly line operations, improving analysis accuracy and efficiency.
Smart Images

Figure 0007588354000001 
Figure 0007588354000002 
Figure 0007588354000003
Abstract
Description
[Technical Field]
[0001] The present disclosure relates to a data analysis device and method for analyzing the efficiency of performing assembly line work. [Background technology]
[0002] Patent Document 1 discloses a production line management device. In a production line where items are weighed and packaged, the production line management device calculates weighing device lost time based on operational information recorded in the operation log, error log, etc., of the packaging device. Weighing device lost time is the time that could have been spent by the packaging device packaging items but that is lost due to a reduction in the packaging device's ability to package items caused by a delay in the operation of the weighing device. In other words, weighing device lost time is a factor in a decrease in the production line's operating rate. The production line management device aims to enable production line managers to easily identify the factors behind the decrease in the production line's operating rate. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2012-252662 Summary of the Invention [Problem to be solved by the invention]
[0004] The present disclosure provides a data analysis device and method that can facilitate analysis of variations in efficiency of performing assembly line work in multiple processes. [Means for solving the problem]
[0005] A data analysis device according to the present disclosure analyzes the efficiency of an assembly line production system that includes multiple processes. The data analysis device includes an acquisition unit that acquires log data indicating a history of the execution of each process each time the assembly line production is performed, and a control unit that generates analysis information indicating the results of an analysis of the history indicated by the log data based on information generated by a probability model that calculates a probability distribution related to the assembly line production. The probability model generates a process efficiency distribution that indicates a probability distribution of the efficiency of each process in the assembly line production, and a variation factor distribution that indicates a probability distribution of factors that cause fluctuations in the assembly line production efficiency at each process, and generates a work efficiency distribution that indicates the probability distribution of the assembly line efficiency based on the process efficiency distribution and the variation factor distribution.
[0006] These general and specific aspects may be realized by a system, a method, and a computer program, as well as combinations thereof. [Effects of the Invention]
[0007] The data analysis device and method disclosed herein can facilitate analysis of fluctuations in efficiency of assembly line operations across multiple processes. [Brief explanation of the drawings]
[0008] [Figure 1] FIG. 1 is a diagram illustrating an overview of a data analysis device according to a first embodiment of the present disclosure. [Figure 2] A block diagram illustrating the configuration of a data analysis device [Figure 3] FIG. 10 is a diagram illustrating an example of the data structure of equipment log data in the data analysis device; [Figure 4] Diagram for explaining the overview of data analysis operations on a manufacturing line [Figure 5] FIG. 10 is a diagram illustrating a line tact estimation model in a data analysis device. [Figure 6] FIG. 10 is a diagram for explaining a tact estimation unit in a line tact estimation model. [Figure 7] 1 is a flowchart illustrating a model construction operation in a data analysis device; [Figure 8]FIG. 10 is a diagram illustrating an example of the data structure of model data in the data analysis device; [Figure 9] 10 is a flowchart illustrating an example of a data analysis operation for a manufacturing line. [Figure 10] Diagram for explaining data analysis operations on a production line [Figure 11] FIG. 10 is a diagram showing an example of a factor evaluation list in a data analysis operation; [Figure 12] FIG. 10 is a diagram for explaining details of processing in data analysis operations. DETAILED DESCRIPTION OF THE INVENTION
[0009] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, more detailed explanation than necessary may be omitted. For example, detailed explanation of well-known matters or redundant explanation of substantially the same configuration may be omitted. This is to avoid unnecessary redundancy in the following explanation and to facilitate understanding by those skilled in the art.
[0010] The applicant provides the accompanying drawings and the following description to enable those skilled in the art to fully understand the present disclosure, and does not intend for them to limit the subject matter described in the claims.
[0011] (Embodiment 1) Hereinafter, a first embodiment of the present disclosure will be described with reference to the drawings.
[0012] 1. Configuration Overview 1 is a diagram illustrating an overview of a data analysis device 2 according to this embodiment. The data analysis device 2 according to this embodiment is applied to performance data management for producing various manufactured products in an assembly line operation in which various pieces of equipment 11 included in a production line 10 of a factory 1 are operated. When the factory 1 is operating, data such as the operation time of each piece of equipment 11 on the production line 10 and the circumstances under which an error occurred is collected and stored as equipment log data D1.
[0013] The manufacturing line 10 illustrated in Fig. 1 includes, for example, n pieces of equipment 11 corresponding to each step in a series of steps from step 1 to step n for manufacturing a finished product, and a work-in-progress buffer 12 for storing unfinished finished products (i.e., work-in-progress) between the pieces of equipment 11 of successive steps. Each piece of equipment 11 is made up of various machines and equipment for carrying out the corresponding step. The work-in-progress buffer 12 is a storage area with a capacity for storing the number of work-in-progress items.
[0014] In order to maintain the production performance of the production line 10 in the factory 1 so as to continuously produce manufactured products, it is necessary to promptly detect the occurrence of a malfunction that reduces production performance and identify the cause of the malfunction. In the production line 10, a situation may arise in which the stoppage of equipment 11 in one process leads to waiting for equipment 11 in another process. With conventional data management technology, it has been difficult to perform data analysis to determine which equipment 11 in all processes of the production line 10 is responsible for the malfunction that reduces performance in the production line 10.
[0015] For example, a decline in production performance in the production line 10 is thought to correspond to an increase in the takt time, i.e., the time it takes to obtain a unit quantity (e.g., one unit) of manufactured products when producing one lot of manufactured products. However, the takt time also increases when equipment with low performance is normally used or when a manufacturing process with high manufacturing difficulty is performed, and such cases are not a malfunction of the production line 10 due to a decline in performance. The data analysis device 2 of this embodiment performs data analysis to distinguish from such cases and detect malfunctions in the production line 10.
[0016] Furthermore, malfunctions in the equipment 11 can be caused by, for example, various types of errors in each piece of equipment 11 causing operation shutdowns. Here, if an error that occurs frequently during normal operation also occurs during malfunctions, the effect of that error on performance degradation is considered to be relatively small. On the other hand, if an error that does not normally occur occurs during malfunctions, the effect of that error is large and it is considered to be highly likely to be the cause of the malfunction. Furthermore, among the various processes in the production line 10, the effect of errors in processes close to the bottleneck is considered to be large. The data analysis device 2 of this embodiment reflects these effects and performs data analysis to evaluate the degree to which each of the various factors is considered to be the cause of the malfunction.
[0017] 1-2.Data analysis device configuration The configuration of the data analysis device 2 in this embodiment will be described with reference to Fig. 2 and Fig. 3. Fig. 2 is a block diagram illustrating the configuration of the data analysis device 2.
[0018] The data analysis device 2 is configured by an information processing device such as a PC. The data analysis device 2 illustrated in Fig. 2 includes a control unit 20, a storage unit 21, an operation unit 22, a display unit 23, a device interface 24, and a network interface 25. Hereinafter, the interface will be abbreviated as "I / F."
[0019] The control unit 20 includes, for example, a CPU or MPU that works in cooperation with software to realize predetermined functions, and controls the overall operation of the data analysis device 2. The control unit 20 reads out data and programs stored in the storage unit 21, performs various arithmetic processing, and realizes various functions.
[0020] For example, the control unit 20 has, as its functional configuration, a calculation unit 30, a model construction unit 31, a detection unit 32, and an evaluation unit 33. The calculation unit 30 calculates various observation values from the equipment log data D1. The model construction unit 31 executes a model construction operation to generate a line tact estimation model 4 (see FIG. 5) described later. The detection unit 32 detects malfunctions in the production line 10 using the line tact estimation model. The evaluation unit 33 evaluates the cause of the malfunction using the line tact estimation model. The operations performed by these various functions of the data analysis device 2 will be described later.
[0021] The control unit 20 executes various programs including commands for implementing the functions of the data analysis device 2, data analysis operations, or model construction operations, such as those described above. The programs may be provided via a communication network such as the Internet, or may be stored on a portable recording medium. The control unit 20 may also be a hardware circuit, such as a dedicated electronic circuit or a reconfigurable electronic circuit, designed to implement the above functions. The control unit 20 may also be configured with various semiconductor integrated circuits, such as a CPU, MPU, GPU, GPGPU, TPU, microcomputer, DSP, FPGA, and ASIC.
[0022] The storage unit 21 is a storage medium that stores programs and data required to realize the functions of the data analysis device 2. As shown in Fig. 2, the storage unit 21 includes a storage unit 21a and a temporary storage unit 21b.
[0023] The storage unit 21a stores parameters, data, control programs, etc. for realizing predetermined functions. The storage unit 21a is configured, for example, with an HDD or SSD. For example, the storage unit 21a stores the above-mentioned programs, equipment log data D1, model data D2, etc.
[0024] The equipment log data D1 is an example of log data that records various information about the history of each process performed in each piece of equipment 11 each time assembly line work is performed on the production line 10. The model data D2 is data that constitutes a line takt estimation model (details will be described later). An example of the equipment log data D1 is shown in FIG. 3.
[0025] The equipment log data D1 includes operation log data D11 and error log data D12, as shown in Fig. 3. The operation log data D11 records, for example, "production product number," "lot number," "line number," "equipment name," "structural parameter s," "structural parameter l," "start time," "end time," "stop time," "production quantity," and "quantity of conforming products" in association with each other. The equipment log data D1 manages the history of assembly line work performed on the production line 10 for each lot, for example, using a "lot number" that identifies the lot of manufactured products.
[0026] For example, in the production of one lot of a manufactured product having a "production part number," it is recorded that each piece of equipment 11 identified by a "equipment name" on a manufacturing line 10 identified by a "line number" operates from a "start time" to an "end time," and is stopped for a "stop time." Furthermore, it is recorded that the number of manufactured products equal to the "production quantity" is obtained at this time, and that of those, the number of good products equal to the "number of good products" is obtained. The structural parameters s and l are set according to the structure of the manufactured product, etc. Furthermore, the error log data D12 in the equipment log data D1 records, for example, for each lot, the "stop time" and the "number of stops" when the equipment 11 with the "equipment name" stops for each "error code" that identifies the type of error.
[0027] 2, the temporary storage unit 21b is configured with a RAM such as a DRAM or an SRAM, and temporarily stores (i.e., holds) data. The temporary storage unit 21b may also function as a work area for the control unit 20, and may be configured with a storage area in the internal memory of the control unit 20.
[0028] The operation unit 22 is a general term for operation members operated by the user. The operation unit 22 may constitute a touch panel together with the display unit 23. The operation unit 22 is not limited to a touch panel, and may be, for example, a keyboard, a touch pad, a button, a switch, or the like. The operation unit 22 is an example of an acquisition unit that acquires various pieces of information input by user operations.
[0029] The display unit 23 is an example of an output unit configured, for example, by a liquid crystal display or an organic EL display. The display unit 23 may display various information such as various icons for operating the operation unit 22 and information input from the operation unit 22.
[0030] The device I / F 24 is a circuit for connecting an external device to the data analysis apparatus 2. The device I / F 24 is an example of a communication unit that communicates in accordance with a predetermined communication standard. The predetermined standard includes USB, HDMI (registered trademark), IEEE1395, WiFi, Bluetooth, etc. The device I / F 24 may constitute an acquisition unit that receives various information from the external device in the data analysis apparatus 2 or an output unit that transmits various information to the external device.
[0031] The network I / F 25 is a circuit for connecting the data analysis apparatus 2 to a communication network via a wireless or wired communication line. The network I / F 25 is an example of a communication unit that performs communication in accordance with a predetermined communication standard. The predetermined communication standard includes communication standards such as IEEE802.3, IEEE802.11a / 11b / 11g / 11ac, etc. The network I / F 25 may constitute an acquisition unit that receives various information or an output unit that transmits various information via the communication network in the data analysis apparatus 2.
[0032] The above-described configuration of the data analysis apparatus 2 is an example, and the configuration of the data analysis apparatus 2 is not limited to this. The data analysis apparatus 2 may be configured with various computers including a server device. The data analysis operation and model construction operation of this embodiment may each be executed by distributed computing. Furthermore, the acquisition unit in the data analysis apparatus 2 may be realized in cooperation with various software in the control unit 20, etc. The acquisition unit in the data analysis apparatus 2 may acquire various pieces of information by reading out the various pieces of information stored in various storage media (e.g., the storage unit 21a) into a working area (e.g., the temporary storage unit 21b) of the control unit 20.
[0033] 2. Operation The operation of the data analysis device 2 configured as above will be described below.
[0034] 2-1. Overall operation The overall operation of the data analysis device 2 in this embodiment will be described with reference to Figures 1 to 4. Figure 4 is a diagram for explaining an outline of the data analysis operation of the data analysis device 2 for the production line 10.
[0035] In this embodiment, for example, the model construction unit 31 of the data analysis device 2 performs a model construction operation to construct a line takt estimation model 4 by machine learning based on the accumulated equipment log data D1. The line takt estimation model 4 is a probabilistic model that estimates the takt time in the production line 10 by Bayesian estimation. The line takt estimation model 4 is constructed so that it can execute estimations regarding various factors that cause malfunctions that degrade performance, and can input observed values regarding the factors as additional knowledge.
[0036] The data analysis device 2 of this embodiment executes a data analysis operation that uses the constructed line tact estimation model 4 to manage the production performance of the production line 10. For example, the detection unit 32 of the data analysis device 2 detects that the production performance has dropped significantly from the result of estimation by the line tact estimation model 4 based on the observed values in the equipment log data D1 that is the subject of the data analysis operation, as a malfunction of the production line 10.
[0037] Fig. 4(A) illustrates an example of an estimation result by the line takt estimation model 4. The horizontal axis of Fig. 4(A) indicates the operation period when the production line 10 was in operation, and the vertical axis indicates the takt time of the production line 10 (similar to Fig. 4(B)). The takt time on the vertical axis is, for example, the effective takt, which will be described later.
[0038] In Figure 4(A), an observation curve Co showing the observed value of the effective tact is shown, along with an estimation curve C1 and confidence interval I1 as the estimation results from the line tact estimation model 4. The estimation curve C1 shows the median value in the probability distribution obtained as the estimation result of the effective tact at each time point by the line tact estimation model 4. The confidence interval I1 indicates an interval defined by a predetermined confidence level in the probability distribution at each time point; for example, the confidence interval I1 with a confidence level of 95% is in the range of 2.5% to 97.5% in the above probability distribution.
[0039] In the example of FIG. 4(A), the observed curve Co significantly exceeds the estimated curve C1 at times T1 and T2, exceeding the upper limit of the confidence interval I1. In other words, at these times T1 and T2, the actual takt time is significantly longer than the result of estimation by the line takt estimation model 4, and it is considered that production performance is declining. The data analysis device 2 of this embodiment detects such times T1 and T2 as a malfunction of the production line 10. For example, with respect to times T1 and T2 when a malfunction of the production line 10 is detected, the evaluation unit 33 of the data analysis device 2 inputs the observed values of various factors that could be the cause of the malfunction into the line takt estimation model 4 and calculates an influence degree that indicates an evaluation value of the influence of each factor on the malfunction of the production line 10.
[0040] Figure 4(B) shows the estimation results by the line tact estimation model 4 when an observation value related to a specific error is input from the state of Figure 4(A). In Figure 4(B), the observation curve Co is the same as in Figure 4(A). On the other hand, the estimation curve C2 and confidence interval I2 change from the estimation curve C1 and confidence interval I1 in Figure 4(A) in response to the line tact estimation model 4 acquiring the input observation value of the error.
[0041] In the example of FIG. 4(B), the observation curve Co falls within the range of the confidence interval I2 at time T2. In other words, it is considered that the estimation result of the line tact estimation model 4, in a state where the observation value of the input error is known, is able to estimate the malfunction at time T2. On the other hand, the observation curve Co remains beyond the upper limit of the confidence interval I2 at time T1. From the above, it can be seen that the influence of the input error is large for the malfunction at time T2 and small for the malfunction at time T1. In the data analysis operation of this embodiment, the line tact estimation model 4 is used to calculate the above-mentioned influence level for each cause such as various errors, thereby quantitatively evaluating the causes of the malfunction.
[0042] In this embodiment, in a production line 10 to which the various data analysis methods described above are applied, a plurality of pieces of equipment 11 are connected in series in the order of processes via in-process buffers 12. Each piece of equipment 11 has its own takt time for executing each process. In addition to error stoppages of individual pieces of equipment 11 in the production line 10, situations such as waiting for a previous process or a subsequent process due to differences in takt times between adjacent processes can occur.
[0043] In light of the above situation, the challenge is to be able to analyze which equipment and what factors are causing the malfunction when the production performance of the entire production line 10 drops. Therefore, in this embodiment, a line takt time estimation model 4 is constructed so that the degree to which errors, etc., of each piece of equipment 11 in each process affect the entire production line 10 is also reflected.
[0044] 2-2. Line tact estimation model The line tact estimation model 4 in this embodiment will be described with reference to FIG.
[0045] 5, the line tact estimation model 4 in this embodiment includes first to n-th tact estimation units 5 provided for each piece of equipment 11 included in a production line 10, and a line integration unit 40 that integrates the estimation results from each tact estimation unit 5. The line tact estimation model 4 in this embodiment corresponds to one production line 10. The data analysis device 2 may construct multiple line tact estimation models 4 for each production line 10.
[0046] The ith takt time estimation unit 5 in the line takt time estimation model 4 is configured as a probabilistic model corresponding to the equipment 11 of the ith process in the production line 10 (i is a natural number from 1 to n). In this embodiment, each takt time estimation unit 5 is constructed from the perspective of decomposing the effective takt time for each equipment 11 into multiple factors, for example, as in the following relational expression (1) (details will be described later). Ot / Pc+Ot(Pc-Gc) / (Pc·Gc)+Et / Gc=Mt / Gc …(1)
[0047] In the above formula (1), Ot is the production time, Pc is the number of manufactured products, Gc is the number of good products, Et is the error stop time, and Mt is the operating time. The above various times and numbers are managed for each lot in the equipment log data D1, for example. For example, the operating time Mt indicates the length of time that a specific piece of equipment 11 operates when one lot of products is manufactured. The production time Ot indicates the length of time it takes for that piece of equipment 11 to execute a process, i.e., for production, during the operating time Mt. The error stop time Et indicates the length of time that production is stopped due to various errors in that piece of equipment 11 during the operating time Mt. The production quantity Pt indicates, for example, the number of products manufactured in one lot. The number of good products Gc indicates the number of good products that are not defective out of the production quantity Pt.
[0048] In the above equation (1), the first term on the left side represents the ideal takt time, the second term represents the defect time width, the third term represents the error time width, and the right side represents the effective takt time. The ideal takt time is the takt time per manufactured product in an ideal case where neither defective products nor stoppages due to errors occur. The ideal takt time corresponds to the efficiency of each process, which varies depending on, for example, the type of manufactured product and various settings of the equipment 11. The defect time width is the time width per good product that is estimated to be the length of time spent manufacturing defective products within the manufacturing time Ot. The error time width is the time width per good product that is calculated by converting the error stoppage time Et into a time width per good product. The effective takt time is the effective takt time per good product that takes into account the effects of defective products and errors.
[0049] Hereinafter, the ideal tact time for the equipment 11 of the i-th process in the manufacturing line 10 is defined as "t0 i " and the defect time width is "y i " and the error time width is "Σ j s j i " and the effective tact is "t1 i " where the error time width Σ j s j i is the error time span s for each error code j i The i-th tact time estimation unit 5 calculates the ideal tact time t0 by Bayesian estimation based on the equipment log data D1 for the equipment 11 of the i-th process. i and the defect time width y i and various error time widths s j i and effective tact t1 i It is constructed to generate probability distributions for each of and (details below).
[0050] In the line tact estimation model 4 of this embodiment, the line integration unit 40 integrates the calculation results by the tact estimation units 5 for each piece of equipment 11 so as to reflect the relationship between the pieces of equipment 11 in the production line 10. For example, it is considered that the slowest tact time among the tact times for each piece of equipment 11 determines the speed of the entire production line 10. From the viewpoint of reflecting this, the ideal tact time t0 and the effective tact time t1 for the entire production line 10 are expressed as in the following equations (2) and (3). t0=max i (t0 i ) …(2) t1=max i (t1 i ) …(3)
[0051] The above formula (2) is the ideal tact time t0 of each equipment 11 in the range of i=1 to n, that is, the 1st process to the nth process in the production line 10. 1 ~t0 n Similarly, for the above formula (3), the effective tact time t1 of each facility 11 is taken as the maximum value. 1 ~t1 n Take the maximum value in
[0052] The line takt time estimation model 4 of this embodiment takes into account error stoppages and the production of defective products as factors that reduce productivity in the entire production line 10, just as it does for each piece of equipment 11. Here, in the production line 10, for example, when equipment 11 in the i-th process stops due to an error, it is considered that the equipment 11 in the subsequent stage can operate without waiting for the amount of work-in-progress that is pre-stored in the in-process buffer 12 between the two. Furthermore, when the equipment 11 in the i-th process stops, the equipment 11 in the preceding stage can operate without waiting as long as the in-process buffer 12 between the two does not reach its capacity.
[0053] As described above, it is considered that the influence of an error stop or the like caused by each piece of equipment 11 in the production line 10 can be mitigated by the in-process buffer 12. From the viewpoint of reflecting the effect of the in-process buffer 12, in this embodiment, the absorption coefficient α i Introduce. t0+Σ iα i (Σ j s j i +y i )=t1…(4)
[0054] In the above equation (4), the absorption coefficient α from i=1 to n in the production line 10 is i Therefore, the error time width Σ j s j i and defect time width y i The weighted sum is calculated by weighting the i-th absorption coefficient α i corresponds to the equipment 11 of the i-th process. The absorption coefficient α i are set to values between 0 and 1, respectively.
[0055] In the line tact estimation model 4 of this embodiment, the larger the capacity of the in-process buffer 12 before and after the equipment 11 of the i-th process, the larger the i-th absorption coefficient α i is closer to "0", and the smaller the tolerance, the lower the absorption coefficient α i is set to "1" in the line integration unit 40. i This makes it possible to reflect in the line tact estimation model 4 that the influence of an error stop or the like in the i-th process on the effective tact t1 of the production line 10 is reduced by the in-process buffer 12.
[0056] The line integration unit 40 of the line tact estimation model 4 calculates the probability distribution of the effective tact t1 of the production line 10, that is, the effective tact distribution D(t1), for example, as in the following equation (5). D(t1)=N(t0+Σ i α i (Σ j s j i +y i ),σ L ) …(5)
[0057] In the above formula (5), N(μ, σ) represents a normal distribution with a mean μ and a standard deviation σ. In this embodiment, the model parameter σ of the effective tact distribution D(t1) of the production line 10 is Lis set as a probability distribution by Bayesian estimation in the line integration unit 40. The effective takt distribution D(t1) is an example of a work efficiency distribution for the effective takt t1 corresponding to the efficiency of the assembly line work in the production line 10.
[0058] The line tact estimation model 4 of this embodiment can use the estimation results of each tact estimation unit 5 in the calculation of the above formula (5), and can also use the observed values of the equipment log data D1 instead of part or all of the estimation results. Note that the function form of the effective tact distribution D(t1) of the production line 10 is not limited to a normal distribution as in formula (5), and may be, for example, a log-normal distribution.
[0059] 2-2-1. About the tact time estimation part The details of each tact estimation unit 5 for each facility 11 in the line tact estimation model 4 of this embodiment will be described with reference to FIG.
[0060] In Figure 6, the i-th tact estimation unit 5 is shown as a DAG (Directed Acyclic Graph). In Figure 6, large black circles indicate observed random variables that are the target of estimation. Open circles indicate unobservable random variables called latent variables, which indicate model parameters. Small black circles indicate parameters that are given as predetermined values, including, for example, hyperparameters that are set in advance before machine learning and parameters that indicate manufacturing conditions.
[0061] Each tact estimation unit 5 includes an ideal tact model 51, an error stop model 52, a defect time model 53, and an effective tact model 54. For example, in the structure parameters s and l that indicate the manufacturing conditions to be estimated in each tact estimation unit 5, the structure parameter s is input to the ideal tact model 51, and the structure parameter l is input to the ideal tact model 51, the error stop model 52, and the defect time model 53.
[0062] In the i-th tact estimation unit 5, the ideal tact model 51 is i For example, the model parameters ws ,w l , σ0 and hyperparameter β0. The ideal tact model 51 has the ideal tact t0 of the i-th process, for example, as shown in the following equation (11): i The probability distribution of the ideal tact distribution D(t0 i ) is calculated. D(t0 i )=N(w s s+w l l,σ0) …(11)
[0063] Ideal tact distribution D(t0 i ) is an example of the process efficiency distribution in this embodiment.
[0064] In this embodiment, the variables s and l representing the structural parameters are categorical discrete values. In the above formula (11), s and l are, for example, vectors with the number of dimensions equal to the number of categories, and only the dimension corresponding to the category to which s and l belong is 1, and the other dimensions are 0. The model parameter w s ,w l is a vector of the same dimension as s and l, and is a random variable that represents the weight for each category of s and l. The model parameter σ0 is the ideal tact t0 i The prior distribution is a random variable that represents the fluctuation of the hyperparameter β0. s ,w l ,σ0 is estimated by Bayesian estimation.
[0065] The error stop model 52 is a model of the error time width s j i The error stopping model 52 is a probabilistic model that estimates the error time width s, and has model parameters γ and θ and hyperparameters γ and θ for each error code, for example. j i The probability distribution of error termination distribution D(s j i ) is calculated. D(s j i )=Exp0(γ,θ) …(12)
[0066] In the above equation (12), Exp0(γ,θ) represents a zero-excessive exponential distribution, which is a combination of the exponential distribution and the Bernoulli distribution, and is defined as the probability density Exp0(x;γ,θ) of the occurrence of value x, as shown in the following equation (13). Exp0(x=0;γ,θ)=θ Exp0(x≠0;γ,θ)=(1-θ)γe -γx …(13)
[0067] The zero-excessive exponential distribution as shown in equation (13) above provides a distribution that includes more cases where x = 0 than the exponential distribution. The model parameters γ and θ for each error code have distinct values depending on, for example, the structural parameter l, and are estimated by machine learning based on the respective hyperparameters γ0 and θ0.
[0068] The defect time model 53 is the defect time width y i The defect time model 53 is a probability model that estimates the defect time width y as shown in the following equation (14), for example, and has model parameters κ and Θ and hyperparameters κ and Θ. i The probability distribution of the failure time distribution D(y i ) is calculated. D(y i )=Γ(κ,Θ) …(14)
[0069] In the above equation (14), Γ(x;κ,Θ) represents a so-called gamma distribution. Each model parameter κ,Θ of the failure time model 53 has a different value depending on, for example, the structural parameter l, and is estimated by machine learning based on the respective hyperparameters κ0,Θ0. Each distribution D(t0 i ),D(y i ),D(s j i ) are examples of the variation factor distribution in this embodiment.
[0070] The effective tact model 54 is the effective tact t1 of the equipment 11. iThe effective tact model 54 of the i-th tact estimation unit 5 is a probability model that estimates the effective tact t1 of the equipment 11 of the i-th process, for example, as shown in the following equation (15): i The effective tact distribution D(t1 i ) is calculated. D(t1 i )=N(t0 i +Σ j s j i +y i ,σ1) …(15)
[0071] According to the above formula (15), for the equipment 11 of the i-th process, the effective tact time t1 i is the ideal tact time t0 i , each error time span s j i and defect time width y i In the i-th tact estimation unit 5, the effective tact model 54 may use the calculation results of the ideal tact model 51, the error stop model 52, and the defect time model 53, or may use the observed values of the equipment log data D1, in the calculation of the above formula (15). The model parameter σ1 of the effective tact model 54 is i It is estimated by machine learning using the distribution due to the hyperparameter β1 as a prior distribution to represent the fluctuations of
[0072] As explained above, the various distributions D(t0 i ),D(s j i ),D(y i ),D(t1 i The function forms of (11), (12), (14), and (15) are merely examples, and are not limited to the respective formulas (11), (12), (14), and (15). For example, function forms of various probability distributions, such as a log-normal distribution, may be adopted as appropriate. Furthermore, although an example has been shown in which the model parameters have different values depending on the structural parameters s and l, other variables, such as the equipment number, may also act on the model parameters.
[0073] 2-3.Model building operation The model construction operation for constructing the above-described line tact estimation model 4 will be described with reference to FIGS.
[0074] 7 is a flowchart illustrating a model construction operation in the data analysis apparatus 2. Each process shown in the flowchart in FIG.
[0075] First, the control unit 20 acquires facility log data D1 to be used for machine learning, for example, from the storage unit 21 (S1). For example, facility log data D1 for the past several months is accumulated in the storage unit 21 in advance and is acquired.
[0076] Next, the control unit 20 functions as the calculation unit 30 and calculates the ideal tact time t0 from the acquired equipment log data D1 as a data set for learning. i , various error time widths s j i , failure time width y i and effective tact t1 i The training data set is made up of, for example, sets of the observed values for each piece of equipment 11 for each lot and the structural parameters s and l, and includes, for example, N sets of the observed values.
[0077] Hereinafter, the superscript "(N)" is attached to the observed values in the learning data set. For example, the calculation unit 30 calculates the ideal tact time t0 of the entire process for each group in the learning data set as shown in equations (2) and (3). i(N) and effective tact t1 i(N) The maximum values in the ideal tact time t0 of production line 10 are (N) and effective tact t1 (N) Thereafter, the control unit 20 functions as the model construction unit 31 (S2 to S8).
[0078] For example, the control unit 20 may calculate the ideal tact time t0 of each process in the learning data set. i(N)Based on the structure parameters s and l, an ideal tact model 51 for each process in the first to nth tact estimation units 5 is constructed (S3). For example, the ideal tact model 51 in the i-th tact estimation unit 5 is constructed based on the observed ideal tact t0 i(N) From the above, the model parameter w that changes for each structural parameter s and l s ,w l , and σ0. These parameter estimates are estimated by Bayesian estimation using the following equation (20): p(w s ,w l ,σ0|t0 i(N) ,s,l) ∝p(t0 i(N) |s,l;w s ,w l ,σ0)p(w s ,w l ,σ0) =N(t0 i(N) ;w s s+w l l,σ0)p(w s ,w l ,σ0) …(20)
[0079] The above equation (20) is based on Bayes' theorem and equation (11). In the above equation (20), the left side is the model parameters w s ,w l ,σ0, and indicates the probability distribution when the observed values for learning are known. (N) ;μ,σ) is the observed value x for a normal distribution with mean μ and standard deviation σ (N) Also, p(w s ,w l ,σ0) is a prior distribution, and generally a distribution that gives a uniform probability for all values in the domain is set.
[0080] In step S3, the control unit 20 applies, for example, the Markov Chain Monte Carlo (MCMC) method to the above equation (20) to calculate the model parameter w that changes for each of the structural parameters s and l. s ,w l, and σ0 are calculated. Each sample group contains M sample values in random order that follow the distribution on the left side of the above formula (20) (see FIG. 8). For example, M is 1,000 to 10,000. The control unit 20 performs the above calculations for the 1st to nth processes to construct an ideal takt model 51 for each process.
[0081] The control unit 20 calculates various error time widths s in the learning data set. j i(N) Based on the structure parameter l, an error stopping model 52 is constructed for each error in each process (S4). The error stopping model 52 for the j-th error code in the i-th process is j i(N) Based on this, the model is constructed by Bayesian estimation using equation (12) instead of equation (11) in equation (20) above. For example, similar to step S3, the control unit 20 calculates a sample group that follows the posterior distribution of the model parameters γ and θ that change for each error code and structure parameter l for each process.
[0082] The control unit 20 calculates the defect time width y i(N) and the structural parameter l, a defect time model 53 for each process is constructed (S5). The defect time model 53 for the i-th process is constructed by Bayesian estimation using equation (14) instead of equation (11) in equation (20) above. For example, similar to step S3, the control unit 20 calculates a sample group that follows the posterior distribution of model parameters κ and Θ that change for each structural parameter l for each process.
[0083] The control unit 20 calculates the effective tact time t1 of each process in the learning data set. i(N)Based on the above, an effective tact model 54 for each process is constructed in the first to n-th tact estimation units 5 (S6). The effective tact model 54 for the i-th process is constructed by Bayesian estimation using equation (15) instead of equation (11) in equation (20) above. For example, similar to step S3, the control unit 20 calculates a sample group that follows the posterior distribution of the model parameter σ1 for each process. At this time, various values in the learning dataset can be used as appropriate.
[0084] Furthermore, the control unit 20 calculates various observed values t0 (N) ,s j i(N) ,y i(N) ,t1 (N) Based on this, the model parameter σ in the line integration unit 40 of the line tact estimation model 4 is calculated. L and each absorption coefficient α i For example, the control unit 20 calculates the model parameter σ by Bayesian estimation using equation (5) instead of equation (11) in equation (20) above. L and each absorption coefficient α i In the same manner as in step S3, a group of samples following the posterior distribution of α is calculated. However, unlike the case of the error stop model 52 for each process, the line takt time estimation model does not take into account the process waiting stop time due to delays in the previous process or the next process as the type of error stop. Furthermore, the control unit 20 calculates each absorption coefficient α i For , the average of the sample values in the sample group is calculated to calculate the coefficient value for each process.
[0085] The control unit 20 calculates the distribution of the various model parameters of the line tact estimation model 4 calculated as above, and the absorption coefficients α i The set value of is stored as model data D2 in the storage unit 21 (S8). An example of the model data D2 is shown in FIG. 8. In FIG. 8, the model parameter w of the ideal tact model 51 of one process in the model data D2 of the line tact estimation model 4 is s ,w l The following shows an example of the state in which the distribution of
[0086] As shown in the column direction of Fig. 8, each model parameter w s ,w l has multiple components that are used depending on the values of the structural parameters s and l. Each model parameter w s ,w l The sample groups are stored, for example, as shown in each row in Fig. 8. The control unit 20 stores the sample values in each sample group in the order generated by, for example, the MCMC method (S8), and randomizes the order of the sample values in each row.
[0087] After storing the model data D2 (S8), the control unit 20 ends the model construction operation of FIG.
[0088] According to the above model construction operation, a line tact estimation model 4 can be constructed by Bayesian estimation using the equipment log data D1 in the production line 10. That is, according to the model construction operation of this embodiment, a model generation method for generating a line tact estimation model 4 is provided.
[0089] 7 may also be applied when reconstructing a constructed line tact estimation model 4 by machine learning of new equipment log data D1. For example, a group of samples of the posterior distributions of the various constructed models 51 to 54 may be used as the prior distributions at the time of reconstruction.
[0090] 2-4. Data analysis operation The data analysis operation of the data analysis device 2 for analyzing the production performance of the manufacturing line 10 using the line tact estimation model 4 constructed as described above will be described with reference to FIGS.
[0091] 9 is a flowchart illustrating an example of a data analysis operation of the production line 10. The process shown in the flowchart of FIG. 9 starts in a state where model data D2 indicating a pre-constructed line tact estimation model 4 is stored in, for example, the storage unit 21. Each process shown in this flowchart is executed by, for example, the control unit 20 of the data analysis device 2.
[0092] First, the control unit 20 of the data analysis device 2 acquires equipment log data D1 to be analyzed using the line tact estimation model 4 (S11). In step S11, for example, several days' worth of equipment log data D1 is acquired. The equipment log data D1 may be acquired from the operation unit 22 by user operation, or may be acquired via the device I / F 24 or the network I / F 25.
[0093] Next, the control unit 20 functions as the calculation unit 30 and calculates various observation values of the analysis target from the acquired equipment log data D1 (S12). Hereinafter, the observation values of the analysis target will be given a superscript "(V)". In step S12, the ideal tact time t0 is calculated as the observation value of the analysis target by the same calculation as in step S2 of FIG. i(V) , various error time widths s j i(V) , failure time width y i(V) and effective tact t1 i(V) The ideal tact time t0 of the production line 10 is calculated. (V) and effective tact t1 (V) is the ideal tact time t0 of the entire process as shown in the above equations (2) and (3). i(V) and effective tact t1 i(V) is obtained as the maximum value at
[0094] In addition, the control unit 20 generates various distributions for each process by Bayesian estimation in the first to n-th tact estimation units 5 in the line tact estimation model 4 according to, for example, the structural parameters s and l in the equipment log data D1 to be analyzed (S13). The control unit 20 calculates the various models 51 to 54 of each tact estimation unit 5 based on the structural parameters s and l and the model data D2, and calculates an ideal tact distribution D(t0 i ), error termination distribution D(s j i ), failure time distribution D(y i ) and the effective tact distribution D(t1 i ) (more details below).
[0095] Furthermore, the control unit 20 generates an effective tact distribution D(t1) of the production line 10 by Bayesian estimation based on various distributions of the first to nth processes in the line integration unit 40 of the line tact estimation model 4 (S14). In step S14, the line integration unit 40 generates an effective tact distribution D(t1) of the production line 10 by Bayesian estimation based on various distributions of the first to nth processes in the model data D2. L and each absorption coefficient α i and the calculation result of step S13, the calculation of equation (5) is performed.
[0096] At this time, the line integration unit 40 calculates the ideal tact distribution D(t0 1 )~D(t0 n ) based on the ideal tact time t0 1 ~t0 n The distribution showing the maximum value of is used as the ideal tact distribution D(t0) of the ideal tact t0 of the production line 10. In addition, the error stop distribution D(s j 1 )~D(s j n ) and the defect time distribution D(y 1 )~D(y n ), the line integrating unit 40 calculates the first to n-th absorption coefficients α i The weighting is performed by the above and is incorporated into the calculation of the effective tact distribution (t1). The details of the processing in steps S13 and S14 will be described later.
[0097] The control unit 20, as the detection unit 32, determines a threshold value for determining a malfunction of the production line 10, that is, a malfunction threshold value th, based on the effective tact distribution D(t1) generated by the line tact estimation model 4 (S15). Fig. 10(A) illustrates an example of the relationship between the malfunction threshold value th and the effective tact distribution D(t1).
[0098] The malfunction threshold th is set such that the probability that the effective cycle t1 exceeds the malfunction threshold th, i.e., the upper probability P(th < D(t1)), in the effective cycle distribution D(t1) is a predetermined ratio. The upper probability P(th < D(t1)) corresponds to the area of the shaded portion in FIG. 10(A). The malfunction threshold th corresponds to the upper limit of the confidence interval in the effective cycle distribution D(t1) (see FIG. 4(A)). The predetermined ratio of the malfunction threshold th is, for example, 5% or the like, and is set complementarily to the confidence level of the confidence interval. In step S15, the control unit 20 determines each malfunction threshold th in each of the effective cycle distributions D(t1) calculated for each of the structural parameters s and l, for example, by counting the sample values in the distribution.
[0099] Next, the control unit 20 functions as the detection unit 32 and detects a malfunction in which the performance of the production line 10 has deteriorated using the calculated malfunction threshold th (S16). An example of the process of step S16 is shown in FIG. 10(B).
[0100] FIG. 10(B) shows an example in which a malfunction is detected in the effective cycle distribution D(t1) of FIG. 10(A). In step S16, the detection unit 32 compares the effective cycle t1 of the observed value in the facility log data D1 to be analyzed (V) with the calculated malfunction threshold, and detects as a malfunction of the production line 10 the case where the effective cycle t1 (V) exceeds the malfunction threshold. Thereby, for example, the malfunction times T1 and T2 in the example of FIG. 4(A) are detected. Note that the detection in step S16 may be performed in units of lots, and for example, a warning display or the like regarding the lot in which a malfunction is detected may be performed.
[0101] The control unit 20, for example, functions as the evaluation unit 33 hereinafter, and performs a process of evaluating the influence degree of various factors when a malfunction of the production line 10 is detected (S17 to S20). The influence degree is calculated based on the amount of information corresponding to the change in the output of the line cycle estimation model 4 before and after inputting the observed value for each factor when a malfunction is detected into the line cycle estimation model 4.
[0102] For example, the ideal cycle t0 of production line 10 during malfunction (V) Based on this, the control unit 20 calculates the influence degree I(t0 (V) ) of the ideal cycle t0 of production line 10 on the malfunction (V) by performing the calculation of the following formula (30) (S17). I(t0 (V) ) = -log[P(t1 (V) <D(t1))] + log[P(t1 (V) <D(t1|t0 (V) ))] …(30)
[0103] In the above formula (30), D(X|Y) represents the distribution of the random variable X under the condition of knowing the information Y. The logarithm log[] is, for example, the binary logarithm. The upper probability P(t1 (V) <D(t1)) in the first term on the right side of the above formula (30) is illustrated in Fig. 10(B). The upper probability P(t1 (V) <D(t1|t0 (V) )) in the second term is illustrated in Fig. 10(C).
[0104] Fig. 10(C) illustrates the effective cycle distribution D(t1|t0 (V) ) under the condition that the ideal cycle t0 of production line 10 of the observed value during malfunction is known from Fig. 10(B). As shown in Figs. 10(B) and (C), the upper probabilities P(t1 (V) ) <D(t1), P(t1 (V) ) <D(t1|t0 (V) ), P(t1 (V) ) <D(t1|t0 (V) ) change according to the knowledge of the ideal cycle t0 of the observed value. According to the above formula (30), the degree to which the situation of malfunction can be explained by the information Y such as the ideal cycle t0 can be calculated as the information amount I(Y) that quantitatively shows this.
[0105] In step S17, the control unit 20 can use, for example, the processing result of step S14 as the effective tact distribution D(t1) of the first term of the above formula (30). For the second term, the control unit 20 can use, for example, the ideal tact t0 of the observed value when the production line 10 is malfunctioning, instead of the ideal tact distribution D(t0) of the production line 10, in the same processing as in step S14. (V) is input to the line tact estimation model 4. As a result, the line tact estimation model 4 performs the same calculation as in step S14 to obtain the ideal tact t0 (V) is the known effective tact distribution D(t1|t0 (V) )
[0106] In addition, the control unit 20 controls various error time widths s j i(V) Based on this, the impact of error stop for each error code in each process I(s j i(V) ) is calculated (S18). For example, the j-th error time width s j i(V) The influence of I(s j i(V) ) is the ideal tact time t0 in equation (30). (V) Instead of assuming that the error time span s j i(V) is known, the calculation can be performed by the same process as in step S17. In this case, the line tact estimation model 4 is calculated by the j-th error stop distribution D(s j i ) instead of the error time width s of the observation value during the malfunction j i(V) By using the corresponding effective tact distribution D(t1|s j i(V) )
[0107] The control unit 20 also determines the failure time width y i(V) Based on this, the impact of defective manufacturing in each process I(y i(V) ) is calculated (S19). For example, the defect time width y i(V) The influence I(y i(V) ) is the ideal tact time t0 in equation (30). (V)Instead of the defect time width y i(V) is known, the calculation can be performed by the same process as in step S17. In this case, the line tact estimation model 4 is based on the defect time distribution D(y i ) instead of the poor time span y i(V) By using the corresponding effective tact distribution D(t1|y i(V) )
[0108] Next, the control unit 20 calculates the various influence degrees I(t0 (V) ),I(s j i(V) ),I(y i(V) ), an evaluation result of each factor is output as a factor evaluation list D5, for example (S20). An example of the factor evaluation list D5 is shown in FIG. 11 as an example of analysis information in this embodiment.
[0109] The factor evaluation list D5 is a list of factors that have a significant impact among various factors that may cause malfunctions in the production line 10. In the example of FIG. 11, the factor evaluation list D5 correlates and displays the "lot number" in which the malfunction was detected, the "equipment" in the production line 10, the "error code," the "stop time," the "number of stoppages," and the "impact" for each error code of each piece of equipment 11. The factor evaluation list D5 makes it possible to compare various error stoppages of multiple pieces of equipment 11 in the production line 10 on the same level.
[0110] In step S20, the control unit 20 extracts a predetermined number of factors (e.g., 10 to 50) in order from the factor with the greatest influence I(Y) to generate a factor evaluation list D5. The influence I(Y) may be negative. This occurs when the factor Y actually improved performance, but overall performance declined due to the deterioration of other factors. Factors Y with negative influence I(Y) values may be deleted from the factor evaluation list D5. If the defective production time or ideal tact time of each piece of equipment 11 is extracted as a factor, the factor evaluation list D5 may display the name of such factor instead of an "error code." The control unit 20 outputs the factor evaluation list D5 to the display unit 23, for example, and displays it. The output of step S20 is not limited to this. For example, the factor evaluation list D5 may be recorded in the memory unit 21 or output to various output units.
[0111] After outputting the factor evaluation list D5 (S20), the control unit 20 ends the data analysis operation of FIG.
[0112] According to the above data analysis operation, the line takt time estimation model 4 constructed by Bayesian estimation can be used to detect malfunctions in the production line 10 and to perform data analysis such as evaluating the influence of various factors in various pieces of equipment 11 that are thought to be the cause of the detected malfunction.
[0113] The details of the processing of steps S13 and S14 in the above data analysis operation will be described below. For example, the control unit 20 performs the processing of steps S13 and S14 by generating (i.e., sampling) samples according to the distribution of the model parameters for each set of each column of the sample group of various model parameters in the model data D2 of Fig. 8.
[0114] For example, in the ideal tact model 51 of the i-th process, the model parameter w s ,w l ,σ0, the ideal tact distribution D(t0 i) is generated (for example, K=1 to 10). The calculation of the model 51 is repeated M times according to the number of columns of the model data D2. As a result, the ideal tact distribution D(t0 i ) results in a set of K×M randomly ordered samples.
[0115] Ideal tact distribution D(t0 i ) sample group is generated by the same calculation as above. i ) is calculated. The ideal tact distribution table D6 is stored in the temporary storage unit 21b, for example, and shows the ideal tact distribution D(t0 i 12B shows an example of a histogram of the sample group of each row in the ideal tact distribution table D6.
[0116] For example, in step S13, the control unit 20 calculates the ideal tact distribution D(t0 i ) sample values. This stores the ideal tact distribution D(t0 i ) can be arranged in a random order. Furthermore, in step S14, for example, the control unit 20 applies equation (2) to each column of the ideal tact distribution table D6 to generate rows containing a sample group of K × M sample values consisting of the maximum values of each column. This makes it possible to obtain the ideal tact distribution D(t0) of the production line 10, as shown in the histogram of FIG. 12(B).
[0117] In step S13, various error stop distributions D(s j i ), and the defect time distribution D(y i ) is also the ideal tact distribution D(t0 i ) are generated as sample groups. In addition, the effective tact distribution D(t1i ) is calculated using various distributions D(t0 i ),D(s j i ),D(y i ) is used. In step S14, for example, the above-mentioned ideal tact distribution D(t0 i ) along with each distribution D(s j i ),D(y i ) are used to calculate a set of samples for each column of the effective takt distribution D(t1) of the production line 10.
[0118] 3. Summary The data analysis device 2 in this embodiment analyzes takt time corresponding to the efficiency of performing an assembly line including a plurality of processes on, for example, a manufacturing line 10. The data analysis device 2 includes various acquisition units 22, 24, 25 and a control unit 20. The acquisition unit acquires equipment log data D1, which is an example of log data showing the history of the execution of each process each time an assembly line is performed (S11). The control unit 20 generates a factor evaluation list D5 as an example of analysis information showing the analysis results of the history shown by the equipment log data D1, based on information generated by a line takt estimation model, which is an example of a probabilistic model that calculates a probability distribution related to assembly line work (S20). The line takt estimation model 4 calculates an ideal takt distribution D(t0 i ), error termination distribution D(s j i ) and the defect time distribution D(y i ) is generated (S13). The ideal takt distribution D(t0 i ) is an example of a process efficiency distribution that shows the probability distribution of the efficiency of each process in an assembly line production. The line tact estimation model 4 uses various distributions D(t0 i ),D(s j i ),D(y i ) based on the effective takt distribution D(t1 i ) is generated (S14).
[0119] The data analysis device 2 described above can estimate the influence of various factors in each process on fluctuations in efficiency in assembly line work, such as a decline in the performance of the production line 10, based on the line takt time estimation model 4. This makes it easier to analyze fluctuations in efficiency in assembly line work. It also improves the analytical accuracy and processing efficiency when a computer performs data analysis on assembly line work as described above.
[0120] In this embodiment, the line tact estimation model 4 uses an absorption coefficient α i Based on this, the error stop distribution D(s j i ) and the defect time distribution D(y i ) is weighted to obtain the effective tact distribution D(t1 i ) is generated (see equations (4) and (5)). This makes it easier to analyze fluctuations in the efficiency of assembly line work, such as performance degradation in the production line 10, by reflecting the situation in which the work-in-process buffer 12 in the production line 10 mitigates the impact of factors such as error stoppages in each process on performance degradation in the production line 10.
[0121] In this embodiment, the control unit 20 calculates the effective tact distribution D(t 1 ) is determined, the efficiency of the assembly line in the history shown by the equipment log data D1 is calculated based on the effective takt distribution D(t 1 The control unit 20 generates analysis information according to the detection result of the performance degradation of the production line 10 (S17 to S20). The effective tact distribution D(t 1 ) can be used to distinguish between usual poor performance and detect a decline in performance of the production line 10. In this way, it is possible to easily analyze fluctuations in the efficiency of assembly line work.
[0122] In this embodiment, the line tact estimation model 4 is a variation factor distribution D(s) related to a specific factor such as a specific error or defective production in a certain process. ji ),D(y i ) instead of the observed value s of the factor in the history shown in the equipment log data D1 j i(V) ,y i(V) Using this, the effective tact distribution D(t1|s j i(V) ),(t1|s j i(V) ) is generated (S18, S19). The analysis information is the effective takt distribution D(t1|s j i(V) ),(t1|s j i(V) ) and the effective takt distribution D(t1) of the unobserved state, the influence I(s j i(V) ),I(y i(V) This makes it possible to evaluate the influence of specific factors in each process when the production line 10 malfunctions, for example, and to facilitate the analysis of the cause of the malfunction in the production line 10.
[0123] In this embodiment, the line tact estimation model 4 generates an effective tact distribution D(t1) of the production line 10 based on the ideal tact distribution D(t0) of the production line 10 (S14). The ideal tact distribution D(t0) of the production line 10 is calculated by dividing the ideal tact distribution D(t0) of each process by the effective tact distribution D(t1) of the production line 10. i ) is the distribution of the maximum value in the ideal tact t0 i Since the longer the time, the lower the efficiency, this is an example of a minimum efficiency distribution that shows the probability distribution of the lowest efficiency in each process. This makes it easier to analyze fluctuations in the efficiency of an assembly line, taking into account the situation in which the maximum takt time, or the minimum efficiency, of each process determines the rate.
[0124] In this embodiment, the line tact estimation model 4 uses the maximum ideal tact t0 of the efficiency of each process in the history indicated by the equipment log data D1 instead of the ideal tact distribution D(t0) of the production line 10. (V)Using the observed efficiency of a specific process, the effective takt distribution D(t1|t0 (V) ) is generated (S17). The analysis information is the effective takt distribution D(t1|t0 (V) ) and the effective takt distribution D(t1) of the unobserved state, the influence I(t0 (V) This makes it possible to evaluate the influence of takt time or the like in a specific process when the production line 10 malfunctions, for example, and makes it easier to analyze the cause of the malfunction in the production line 10.
[0125] In this embodiment, the assembly line is a task of producing manufactured products, which are an example of goods, on a production line 10, which is an example of a production line including a plurality of pieces of equipment that perform each process. The causes of each process include both a stoppage due to an error in the equipment that performs the process and the occurrence of defective products. For example, in a situation where the impact of either a stoppage due to an error or the occurrence of defective products can be ignored, it is not necessary to include both of the above in the causes of each process.
[0126] The data analysis device 2 in this embodiment further includes a storage unit 21 that stores model data representing the line tact estimation model 4. The control unit 20 executes calculations to generate an effective tact distribution D)(t1) as the line tact estimation model 4 based on the model data D2 stored in the storage unit 21 (S13, S14). Executing calculations for the line tact estimation model 4 makes it easier to analyze fluctuations in the efficiency of assembly line work.
[0127] In this embodiment, the line tact estimation model 4 is constructed by Bayesian estimation based on the history of past assembly line work (S1 to S8). By using Bayesian estimation, various model parameters in the line tact estimation model 4 can be obtained as probability distributions, making it easier to perform various analyses.
[0128] The data analysis method in this embodiment is a method for analyzing the efficiency of an assembly line including multiple processes. This method includes the steps of: acquiring equipment log data D1 indicating the execution history of each process each time the assembly line is performed (S11); and generating analysis information indicating the results of analyzing the history indicated by the equipment log data D1 based on information generated by a line takt estimation model 4 that calculates a probability distribution related to the assembly line (S16-S20). The line takt estimation model 4 generates a process efficiency distribution indicating the probability distribution of the efficiency of each process in the assembly line, and a variation factor distribution indicating the probability distribution of factors that cause fluctuations in the assembly line efficiency at each process (S13), and generates a work efficiency distribution indicating the probability distribution of the assembly line efficiency based on the process efficiency distribution and the variation factor distribution (S14).
[0129] In this embodiment, a program for causing a computer to execute the above-described data analysis method may be provided. The data analysis method of this embodiment makes it easy to analyze fluctuations in efficiency of assembly line work performed in multiple processes.
[0130] (Other embodiments) As described above, embodiment 1 has been described as an example of the technology disclosed in this application. However, the technology in this disclosure is not limited to this, and can be applied to embodiments in which appropriate modifications, substitutions, additions, omissions, etc. are made. Furthermore, it is also possible to combine the components described in each of the above embodiments to create a new embodiment. Therefore, other embodiments will be described below as examples.
[0131] In the above-described first embodiment, the data analysis apparatus 2 that performs both a data analysis operation and a model construction operation has been described. In this embodiment, for example, the model construction operation may be performed by an information processing apparatus external to the data analysis apparatus 2. The data analysis apparatus 2 of this embodiment may acquire model data D2 generated externally via, for example, various I / Fs 24 and 25, and perform the same data analysis operation as in the above-described embodiment.
[0132] In each of the above embodiments, an example of an operation in which calculations of the line tact estimation model 4 are executed in the data analysis device 2 has been described. In this embodiment, calculations of the line tact estimation model 4 may be executed outside the data analysis device 2. The data analysis device 2 of this embodiment may acquire information generated in the external line tact estimation model 4 by transmitting the information to an information processing device or the like having the external line tact estimation model 4 via, for example, various I / Fs 24 and 25. Even in this case, the control unit 20 of the data analysis device 2 can generate analysis information based on the acquired information in the same way as in each of the above embodiments.
[0133] In addition, in each of the above embodiments, the factor evaluation list D5 is given as an example of the analysis information, but the analysis information in this embodiment is not limited to this. For example, the control unit 20 of the data analysis device 2 may output various warnings as analysis information when a malfunction is detected in the data analysis operation of FIG. 9 (S16), or may generate a graph display such as that of FIG.
[0134] Furthermore, while the above embodiments have described an example of the line tact estimation model 4, the line tact estimation model 4 in this embodiment is not limited to this and may be modified in various ways. For example, the line tact estimation model 4 in this embodiment may be configured to allow an equipment number to be input as a parameter of an explanatory variable in addition to or instead of the structural parameters s and l. For example, an equipment number identifying multiple pieces of equipment included in different production lines 10 in a common process may be used for each process. Furthermore, while normal distributions and gamma distributions are used as the ideal tact and the distributions of each factor, these are not limiting. A distribution that best fits the actual distribution of the observed values should be selected. Furthermore, in the line tact estimation model 4 in this embodiment, the effective tact model 54 for each process may be omitted.
[0135] In each of the above embodiments, the data analysis device 2 analyzes a malfunction that reduces the production performance of the production line 10, i.e., a decrease in the efficiency of the assembly line. In this embodiment, the data analysis device 2 is not limited to analyzing a malfunction of the production line 10, but may also analyze a good performance that reduces production performance, or may analyze various fluctuations including an increase in the efficiency of the assembly line.
[0136] Furthermore, in each of the above embodiments, an example has been described in which the data analysis device 2 and the data analysis method are applied to the production line 10 in the factory 1, but the present invention is not limited to this and various applications are possible. For example, the production line 10 to which the data analysis device and method of the present embodiment are applied may be configured across multiple factories, rather than being limited to one factory 1.
[0137] Furthermore, the data analysis device and method of this embodiment may be applied not only to the manufacturing line 10 but also to a production line using multiple pieces of equipment in an assembly line including various processes. For example, the processes in the production line are not limited to manufacturing, but may also include various processes such as inspection and packaging. Even in such cases, equipment stoppages or the occurrence of defective products can result in a decline in production performance. Therefore, in this embodiment, the same data analysis method as above can be used to easily analyze fluctuations in assembly line efficiency.
[0138] Furthermore, the data analysis device and method of this embodiment can be applied to assembly line work using various equipment, not limited to machinery and equipment in a factory. For example, the assembly line work to be analyzed in this embodiment may be delivery or data communication, etc. The equipment used in the assembly line may be a delivery vehicle or communication equipment, etc. Furthermore, the data analysis device and method of this embodiment are not necessarily limited to assembly line work using equipment, but can also be applied to various manual tasks. Even in this case, fluctuations in the efficiency of the assembly line can be analyzed based on log data of the assembly line's history, as in the above embodiments.
[0139] As described above, the embodiments have been described as examples of the technology in the present disclosure, and for that purpose, the accompanying drawings and detailed description have been provided.
[0140] Therefore, the components shown in the accompanying drawings and detailed description may include not only essential components for solving the problem, but also components that are not essential for solving the problem in order to illustrate the above technology. Therefore, the fact that these non-essential components are shown in the accompanying drawings or detailed description should not be interpreted as immediately indicating that these non-essential components are essential.
[0141] Furthermore, since the above-described embodiments are intended to illustrate the technology of the present disclosure, various modifications, substitutions, additions, omissions, etc. may be made within the scope of the claims or their equivalents. [Industrial Applicability]
[0142] The present disclosure is applicable to the analysis of efficiency when performing various assembly line operations, and is applicable, for example, to the analysis of production performance of assembly line operations that use multiple pieces of equipment on a production line.
Claims
1. A data analysis device for analyzing efficiency of an assembly line including a plurality of processes, an acquisition unit that acquires log data indicating a history of each process performed each time the assembly line is performed; a control unit that generates analysis information indicating an analysis result of a history indicated by the log data based on information generated by a probability model that calculates a probability distribution regarding the assembly line; The probabilistic model is generating a process efficiency distribution indicating a probability distribution of the efficiency of each process in the assembly line, and a variation factor distribution indicating a probability distribution of factors that vary the efficiency of the assembly line in each process; generating a work efficiency distribution indicating a probability distribution of the efficiency of the assembly line based on the process efficiency distribution and the variation factor distribution; the probabilistic model uses an observed value of a specific factor in the history indicated by the log data instead of the variation factor distribution for the specific factor to generate a work efficiency distribution in a state in which the specific factor is observed; The analysis information includes information for evaluating the influence of the factor on the efficiency of the assembly line according to a difference between a work efficiency distribution in a state where the factor is observed and a work efficiency distribution in a state where the factor is not observed. Data analysis equipment.
2. The probability model generates the work efficiency distribution by weighting the variation factor distribution for each of the processes based on a coefficient preset for each of the processes. The data analysis apparatus according to claim 1 .
3. The control unit is detecting a fluctuation in efficiency based on a probability that the efficiency of the assembly line in the history indicated by the log data occurs in the work efficiency distribution; generating said analysis information in response to said detection result of said efficiency variation; The data analysis device according to claim 1 or 2.
4. The probability model generates the work efficiency distribution based on a minimum efficiency distribution that indicates a probability distribution of the lowest efficiency among the efficiencies of the plurality of processes. The data analysis device according to any one of claims 1 to 3.
5. the probabilistic model uses an observed value of the efficiency of a specific process among the efficiencies of the processes in the history indicated by the log data instead of the minimum efficiency distribution to generate an operation efficiency distribution in a state in which the efficiency of the process is observed; The analysis information includes information for evaluating the influence of the efficiency of the process on the efficiency of the assembly line according to a difference between a work efficiency distribution in a state where the efficiency of the process is observed and a work efficiency distribution in a state where the efficiency of the process is not observed. The data analysis apparatus according to claim 4.
6. The assembly line production is a process of producing an article on a production line including a plurality of pieces of equipment that perform each of the steps, The cause of each process includes at least one of a stop due to an error in equipment that performs the process and the occurrence of a defective product in the article. The data analysis device according to any one of claims 1 to 5.
7. A storage unit for storing model data representing the probabilistic model, The control unit executes a calculation to generate the work efficiency distribution as the probabilistic model based on the model data stored in the storage unit. The data analysis device according to any one of claims 1 to 6.
8. The probabilistic model is constructed by Bayesian estimation based on the history of the assembly line operations. The data analysis device according to any one of claims 1 to 7.
9. A data analysis method for analyzing efficiency of an assembly line including a plurality of processes, comprising: acquiring log data indicating a history of each process performed each time the assembly line is performed; generating analysis information indicating an analysis result of the history indicated by the log data based on information generated by a probability model that calculates a probability distribution regarding the assembly line; The probabilistic model is generating a process efficiency distribution indicating a probability distribution of the efficiency of each process in the assembly line, and a variation factor distribution indicating a probability distribution of factors that vary the efficiency of the assembly line in each process; generating a work efficiency distribution indicating a probability distribution of the efficiency of the assembly line based on the process efficiency distribution and the variation factor distribution; the probabilistic model uses an observed value of a specific factor in the history indicated by the log data instead of the variation factor distribution for the specific factor to generate a work efficiency distribution in a state in which the specific factor is observed; The analysis information includes information for evaluating the influence of the factor on the efficiency of the assembly line according to a difference between a work efficiency distribution in a state where the factor is observed and a work efficiency distribution in a state where the factor is not observed. Data analysis methods.
10. A program for causing a computer to execute the data analysis method according to claim 9.
Citation Information
Patent Citations
Production line management device
JP2012252662A
Fault diagnosis apparatus, fault diagnosis system, fault diagnosis method, and program
JP2017167599A
Production management device, production management system, and production management method
JP2020027470A
Manufacturing management device, manufacturing system, and manufacturing management method
WO2019013196A1
Cited By
JPWO2023233745A1
