JP2020113165A - Failure prediction system - Google Patents

Abstract

To provide a failure prediction system capable of highly accurately predicting failures.SOLUTION: A failure prediction system comprises: a sales information storage part that is managed by a seller selling a device or a component to a user who uses the device or the component and that records sales information of the device or the component for the user; a collection part that collects operation information of the device or the component via a network; an operation information storage part that records the operation information of the device or the component in association with the sales information stored in the sales information storage part; and a prediction part that predicts a failure time point of the device or the component by performing machine learning on the basis of the sales information and the operation information.SELECTED DRAWING: Figure 2

Description

The present invention relates to a device or component failure prediction system.

Conventionally, for example, Patent Document 1 discloses a failure prediction notification system for continuing the operation without stopping the apparatus.

JP, 2009-217770, A

For example, a welding or processing system as an example of an apparatus includes a welding or processing apparatus, a robot, or the like. Welding or processing systems consist of multiple components, such as parts that are replaced or maintained due to failure. Generally, a user of a welding or processing system purchases a device or part from a distributor of the device or part or requests maintenance. Such sales and maintenance history is managed by a customer relationship management (CRM) system that uses cloud computing or the like for each user, and the dealer uses it for managing and analyzing customer information.

The above-mentioned conventional failure prediction notification system predicts failures from the user's point of view, and there is no known method of predicting failures from the point of view of a store serving the user and utilizing the information. ..

The present invention has been made in consideration of such circumstances, and an object thereof is to provide a failure prediction system capable of highly accurate failure prediction.

In order to solve the above problems, the failure prediction system according to the present invention is managed by a seller who sells the device or part to a user who uses the device or part, and sales information of the device or part to the user. Associated with the sales information recorded in the sales information storage unit, a collection unit that stores the operation information of the device or the component via a network, and the operation information of the device or the component And an operation information storage section for recording the operation information, and a prediction section for predicting a failure time point of the device or component by machine learning based on the sales information and the operation information.

In the failure prediction system according to the present invention, highly accurate failure prediction can be performed.

The schematic functional block diagram which shows the functional structure of the failure prediction system in this embodiment. The flowchart explaining the failure prediction process performed by the failure prediction system in this embodiment. FIG. 5 is a sequence diagram particularly explaining a process in the manufacturing line and the failure prediction system. Flowchart explaining substitute product proposal processing executed by the failure prediction system

An embodiment of a failure prediction system according to the present invention will be described with reference to the accompanying drawings. In the present embodiment, the case where the failure prediction system according to the present invention is applied to the failure prediction of welding or a processing system, which is used for welding for arc welding and various processing/forming, will be described as an example. Hereinafter, the welding or processing system is simply referred to as "welding system". Further, when simply referred to as welding, it may mean “welding or processing”.

FIG. 1 is a schematic functional block diagram showing a functional configuration of a failure prediction system 1 according to this embodiment.

In the following description, the “user” refers to a person who operates and uses the apparatus or parts in the user equipment including the welding system 35. A "seller" is a person who sells a device or a component to a user, maintains the device or the component, or modifies the device or the component. The “maker” refers to a person who manufactures devices and parts that the seller sells to the user and sells to the seller. The "apparatus" and the "parts" are the welding system 35, the robot body 36, the jigs/sensors 37, the robot controller 38, the arithmetic CPU 39, or these parts, the PLC 32, the PLC-GW 33, and the like, which are included in the cell 31. It includes the communication GW 34 or parts included therein.

The failure prediction system 1 is connected to the network 2. The failure prediction system 1 is connected to the manufacturing lines 3,... 3n, the seller terminal 4, the maker terminal 5, and the user terminals 6,..., 6n via the network 2.

The production lines 3,... 3n are facilities managed by a user and composed of a plurality of cells. A plurality of (different users) user equipments managed by a plurality of users are connected to the network 2. However, since each user equipment has almost the same configuration, only one user equipment is illustrated here. explain. Further, since the respective production lines 3,... 3n have substantially the same configuration, only one production line 3 will be described in detail here.

The production line 3 has a cell 31, a PLC 32, a PLC-GW 33, and a communication GW 34.

The cell 31 has a welding system 35, a robot body 36, jigs/sensors 37, and a robot controller 38. The cell 31 is a concept in which a small section included in the manufacturing line 3 is used as a unit, and as shown in FIG. 1, the same manufacturing line 3 may include a plurality of cells 31a, 31b, 31c.

The welding system 35 includes a welder, a feeder control box, a wire feeder, a welding torch, a torch cable, and the like. The robot body 36 is a robot for automatically performing welding. The jig/sensors 37 include a jig such as a positioner, a position sensor, a temperature sensor, sensors such as a vibrometer, and an imaging device.

The welding system 35 and the robot body 36 are connected to a robot controller 38. The robot controller 38 controls the welding system 35 and the robot body 36 based on the control of a PLC (Programmabl Logic Controller) 32.

The PLC 32 is connected to the robot controller 38 and the jigs/sensors 37, and controls them based on pre-programmed control contents, whereby the welding system 35, the robot main body 36, and the jigs/sensors 37 (cell 31). ) Is controlled at a higher level.

The welding system 35, the robot controller 38, and the PLC 32 are connected to the arithmetic CPU 39. An arithmetic CPU (Central Processing Unit) 39 acquires physical quantities relating to various weldings from the welding system 35, the robot controller 38, and the PLC 32.

The physical quantity relating to welding corresponds to all the quantifiable information obtained from the production line 3. The physical quantity related to welding includes, for example, operation information of a motor that drives the axis of the robot body obtained from the robot controller 38, or welding conditions obtained from a welding device. The motor operation information includes, for example, a motor current command value, an actual current value, a motor speed command value, an actual speed, or encoder position information. Welding conditions include, for example, welding method, welding current, welding voltage, welding wire feed rate, welding speed, welding waveform adjustment amount, protrusion amount, welding torch advance/reverse angle, aiming angle, aiming position, shield gas flow rate. , Weaving conditions, arc sensor conditions, and welding position offset amount during multi-layer welding. Further, the physical quantity relating to welding includes various values measured by the welding system 35 and jigs/sensors 37 that operate based on these welding conditions. The physical quantity of the welding condition is measured by a predetermined measuring device.

In addition, the physical quantity relating to welding is, for example, imaging data of a welding portion imaged by an imaging device, appearance of a welding bead obtained by processing this imaging data, extra height of bead, bead width, and spatter generation amount. Including. Further, the physical quantity relating to welding includes the amount of penetration obtained from the penetration measuring device and the arc sound waveform obtained from the sound collector.

The PLC 32a is connected to a plurality of cells 31a and 31b as shown in FIG. 1 as an example. For example, a different PLC 32b is provided in the different cell 31c, and this PLC 32b is also connected to the cell 31c having a configuration similar to that of the cell 31a.

The PLC 32 and the arithmetic CPU 39 are sequentially connected to a PLC-GW (PLC-Gateway) 33 and a communication GW (communication Gateway) 34. The PLC-GW 33 converts the communication protocol of the plurality of PLCs 32a and 32b connected to the device and the arithmetic CPU 39 into a predetermined format that can be used in the failure prediction system 1. The PLC-GW 33 transmits the physical quantity related to the welding obtained from the CPU 39 for calculation to the failure prediction system 1 via the communication GW 34 and the network 2. The PLC-GW 33 acquires welding conditions and the like at regular intervals and transmits them. At this time, the operation information regarding the PLC-GW 33 or the communication GW 34 may be transmitted to the failure prediction system 1 together with the physical quantity regarding welding.

A unique ID is assigned to each device that provides information to the failure prediction system 1 such as the welding system 35 and the jigs/sensors 37. Further, components (welding torch, welding wire, welding tip, etc.) that configure the welding system 35 and jigs/sensors 37 (hereinafter, sometimes simply referred to as “device”), and elements configured by devices and components. Similarly, a unique ID is attached to (for example, the axis of the robot body). The information provided to the failure prediction system 1 is transmitted in an identifiable manner in association with the ID.

The process from the communication GW 34 to the transmission to the network 2 is performed by the user of the welding system 35, and after being transmitted to the failure prediction system 1, is performed by the seller.

The seller terminal 4 is a terminal (computer) used by the seller. The seller uses the seller terminal 4 to access the failure prediction system 1 or receive a notification from the failure prediction system 1.

The manufacturer terminal 5 is a terminal used by the manufacturer. The manufacturer uses the manufacturer terminal 5 to access the information in the failure prediction system 1 that is the subject of disclosure to the manufacturer terminal 5, and receives a notification from the failure prediction system 1.

The user terminals 6,..., 6n are terminals used by each user. 6n of a plurality of (different users) managed by a plurality of users are connected to the network 2. However, since each user terminal 6,... 6n has substantially the same configuration, the user terminal 6 As described below. The user uses the user terminal 6 to access the information in the failure prediction system 1 that is the subject of disclosure to the user terminal 6, and receives a notification from the failure prediction system 1. Further, the user terminal 6 can access, for example, a website managed by the seller, and can place an order for a device or a part through the website.

The failure prediction system 1 is a system mainly used by sellers, for example, a system using SaaS (Software as a Service) using cloud computing. Further, the failure prediction system 1 is a customer relationship management (CRM) system for a seller, and is used for managing and analyzing customer information.

The failure prediction system 1 includes a collection unit 11, a storage unit 12, a calculation unit 13, a notification unit 14, and an ordering unit 15.

The collection unit 11 acquires a physical quantity related to welding from the manufacturing line 3 via the network 2. The collection unit 11 classifies the acquired physical quantities related to welding, etc., according to the given IDs, and aggregates them in time series for each unit process of the manufacturing line to generate information corresponding to the operation cycle of each device and part. To do. The collection unit 11 records this information in the operation information storage unit 18 as operation information.

The storage unit 12 includes a sales information storage unit 17 and an operation information storage unit 18.

The sales information storage unit 17 records the sales information of the device or parts to the user. The sales information may include a sales history of products or parts made by the seller to the user, a maintenance history of the device or parts, or a modification history of the device or parts. The sales information storage unit 17 has, for example, a tree structure with user information (user name, etc.) as the top. For example, the sales information storage unit 17 sequentially stores the information about the manufacturing line 3 (user equipment), the information about the cell 31, the information about the device included in the cell 31, and the information about the elements or parts included in the device under the user information. I am recording. The sales information storage unit 17 records the information with the above-mentioned unique ID.

The sales information storage unit 17 acquires, from the seller terminal 4, information regarding sales, maintenance, and modification performed by the seller to the user, and records the information. In addition to the sales information, the sales information storage unit 17 holds information necessary for the seller to sell, such as the required inventory quantity of each device or component for each user. The sales information storage unit 17 also holds substitute item information regarding substitute items for each device or part. These pieces of information are appropriately transmitted from the seller terminal 4 and recorded (updated, added, or corrected) in the sales information storage unit 17.

The operation information storage unit 18 records the operation information obtained from the collection unit 11. The operation information is recorded in association with the sales information recorded in the sales information storage unit 17. The association is performed by the above-mentioned ID.

The calculation unit 13 includes a prediction unit 19 and a suggestion unit 20.

The predicting unit 19 predicts a failure timing (failure prediction time) of the device or the component by machine learning based on the sales information and the operation information. Specifically, the prediction unit 19 machine-learns past sales information and operation information accumulated in the sales information storage unit 17 and the operation information storage unit 18 from the time the device operates until the time the device fails, and Generate an inference model to infer the point of failure of the part. For example, the prediction unit 19 qualitatively (probabilistically) evaluates the change in the operation information up to the time of the failure, and performs machine learning. The prediction unit 19 obtains a difference from the obtained inference model and the operating state until the current device or component fails, and obtains a curve (transition) up to the failure prediction time. The predicting unit 19 updates the estimation model each time the past sales information and operation information from when the device is operated until it breaks down is obtained, and further accumulates information about the manufacturing lines of other users to improve accuracy. It is designed to predict high failure points.

For example, the predicting unit 19 machine-learns the operating information about the motor regarding the failure prediction of the motor of the robot main body 36 in terms of the cycle until the failure, slows the responsiveness, changes in the load factor, and ambient temperature and vibration as additional information. Generates an inference model that considers the effect of the frequency on the failure. For machine learning, a method such as deep learning can be used, and various methods such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, transduction, and multitask learning can be applied. The same applies to the proposing unit 20.

Here, “fault” means a state in which the device or part cannot be used for welding, and includes a state in which replacement with a new device or part is necessary. “Failure” also includes a condition in which a device or part can be used for welding but is unable to obtain the required weld quality.

The suggestion unit 20 proposes a substitute for the device or the component by machine learning based on the sales information, the operation information, and the component information. Specifically, the suggestion unit 20 machine-learns past sales information and operation information accumulated in the sales information storage unit 17 and the operation information storage unit 18, and substitutes the currently used device or part for a substitute. An inference model is generated to evaluate the case. The proposing unit 20 determines whether or not there is a preferred alternative product over the product or part currently in use, based on the evaluation of the alternative product obtained from this estimation model.

The notification unit 14 notifies the user terminal 6 based on the estimation results of the prediction unit 19 and the suggestion unit 20. For example, when the time until the failure prediction time is less than the notification time that is the preset notification time, the notification unit 14 notifies the user terminal 6 by e-mail or the like. In addition, the notification unit 14 notifies the user terminal 6 by e-mail or the like when there is a substitute product to be proposed to the user.

The ordering unit 15 automatically performs an ordering process for a device or a component whose failure is estimated based on the estimation result of the prediction unit 19. For example, when the time up to the failure prediction time is less than the preset ordering time, which is the time for placing an order, the ordering unit 15 records the information of the relevant part in the sales information storage unit 17, and sells the content. To the person terminal 4. The seller ships the device or part to the user based on this notification.

The failure prediction system 1 as described above includes detailed information regarding the devices or parts used in the manufacturing line 3 already recorded in the sales information storage unit 17, information regarding the manufacturing line, and the operation obtained from the manufacturing line 3. Record in association with information. For this reason, it is possible to perform machine learning by more reflecting the usage environment of the user, as compared with machine learning using only operation information obtained from the device or component.

Next, the processing executed by the failure prediction system 1 in this embodiment will be described in detail.

FIG. 2 is a flowchart illustrating a failure prediction process executed by the failure prediction system 1 according to this embodiment.

FIG. 3 is a sequence diagram particularly explaining the process in the manufacturing line 3 and the failure prediction system 1.

In step S1 of FIG. 2, the collection unit 11 acquires operation information. That is, the collection unit 11 acquires the physical quantity related to welding (step S11 in FIG. 3) acquired by the manufacturing line 3 from the device or the component via the network 2 (step S12). The collection unit 11 acquires operation information by performing the above-described required processing on the physical quantity related to welding (step S13).

In step S2, the operation information storage unit 18 acquires and records the operation information from the collection unit 11 (step S14). At this time, the operation information storage unit 18 records in association with the sales information stored in the sales information storage unit 17 (step S15).

In step S3, the prediction unit 19 acquires operation information from the operation information storage unit 18 (step S16). In addition, the prediction unit 19 acquires sales information from the sales information storage unit 17 (step S17). The prediction unit 19 performs machine learning based on the acquired information and updates the inference model for predicting a failure (step S18). The estimated model may be updated at various timings, such as each time new sales information is recorded in the sales information storage unit 17.

In step S4, the prediction unit 19 acquires the failure prediction time point based on the estimation model (step S19). The prediction unit 19 outputs the acquired failure prediction time point to the notification unit 14 and the ordering unit 15 (steps S20 and S21).

In step S5, the notification unit 14 determines whether the failure prediction time is less than the preset notification time. When the notification unit 14 determines that the time is less than the notification time (YES in step S5), the time until the time when the device or the component is predicted to fail in the user terminal 6 in step S6 corresponds to the notification time. It is notified that it is less than the time (step S22). By receiving this notification, the user can perform necessary maintenance and ordering of replacement parts. As a result, the downtime due to an unintended failure can be reduced.

In step S7, the ordering unit 15 determines whether or not the ordering time is shorter than a preset ordering time until the failure prediction time. If the ordering unit 15 determines that it is less than the ordering time (YES in step S7), in step S8, the ordering unit 15 performs an ordering process for a device or a component that needs to be replaced due to a failure (step S23). This process is performed by the failure prediction system 1 automatically determining the required device or component without the user performing the ordering process. The ordering unit 15 can also determine the number of orders by referring to the user's required inventory quantity recorded in the sales information storage section 17. As a result, the user can save time and effort for ordering work, and inventory management can be automated. Further, the seller can also save the trouble of interacting with the user. When the notification unit 14 determines that the time is not less than the notification time (NO in step S5), the ordering unit 15 determines that the time is not less than the order time (NO in step S7), and after S8, the process returns to step S1. This process is repeatedly executed while the manufacturing line 3 is in operation.

Next, the alternative product proposing process executed by the failure prediction system 1 will be described.

FIG. 4 is a flowchart for explaining the alternative item proposing process executed by the failure prediction system 1. This substitute item proposing process may be performed at a constant cycle or may be performed at a predetermined timing (for example, a timing of a device or component failure). Although the process corresponding to the alternative product proposing process is described following the sequence diagram of FIG. 3 used for the description of the failure prediction process described above, the timing at which the process is executed is not limited to this.

In step S31, the suggestion unit 20 appropriately acquires the sales information and the substitute item information from the sales information storage unit 17 (step S41 in FIG. 3). The sales information and the substitute information are appropriately input from the seller terminal 4 and recorded in the sales information storage unit 17 (step S42 in FIG. 3).

In step S32, the proposing unit 20 performs machine learning based on the acquired information, and updates the estimation model for estimating the failure prediction time point according to the device or component (step S44). In step S33, the proposing unit 20 evaluates when a substitute product is used, based on the estimation model (step S45). The estimated model may be updated at various timings, such as each time new sales information is recorded in the sales information storage unit 17.

As an example, the evaluation of the welding tip can be evaluated by the wear that can be judged from the welding current and the welding voltage. The proposing unit 20 selects a welding tip as an alternative product that reduces wear and improves productivity based on the estimation model. The proposing unit 20 calculates, for example, the cost of the welding tip in a certain period from the exchange cycle and the price of the welding tip currently used. Further, the proposing unit 20 calculates the cost of the welding tip in a certain period from the welding tip replacement cycle and the component price predicted when the welding tip as the substitute is used. The proposing unit 20 can compare these costs, and if the cost of using the substitute is small, it can evaluate that the substitute should be used.

Moreover, as another example, the welding wire can be evaluated by the wire resistance that can be determined from the current and voltage of the wire wire motor. The proposing unit 20 selects a welding wire as an alternative to reduce the wire resistance based on the estimation model. The proposing unit 20 evaluates, for example, the replacement frequency, the yield, and the number of times of stoppage or idling (so-called chocolate stop) due to a primary trouble caused by the wire as the evaluation item, the welding wire currently used and the welding wire as a substitute. To compare. The proposing unit 20 can evaluate that the substitute should be used if the substitute is better evaluated.

In step S34, the proposing unit 20 determines whether the evaluation is improved when the substitute is used as compared with the case where the currently used device or component is used. When it is determined that the improvement is made (YES in step S34), the suggestion unit 20 outputs the evaluation information to the notification unit 14 (step S46). In step S35, the notification unit 14 notifies the user terminal 6 of the content of proposing a substitute product based on the evaluation information (step S47). On the other hand, when the proposing unit 20 determines that there is no improvement (NO in step S34), the process ends.

Such a failure prediction system 1 stores the operation information acquired from the manufacturing line 3 in association with a system such as a CRM system that is managed by a seller and holds customer information and sales information. As a result, the seller can obtain information associated with the sales information, sales history, maintenance history, or modification history owned by the seller and the operation information without the trouble of inputting, collecting device information, or inputting information. be able to. The failure prediction system 1 can perform machine learning based on this information to perform more accurate failure prediction that is more realistic.

In addition, the failure prediction system 1 enables the seller to obtain information about failure prediction, and therefore, the failure prediction system 1 can be used for the sales prediction of the seller itself and the manufacturing prediction and sales prediction of a manufacturer who sells products and parts to the seller. Information can be utilized. As a result, the seller or the maker can enjoy the merit of being able to predict the proper supply timing and supply quantity of the product or component and also to propose the supply before the stock runs out. Further, the manufacturer can quantitatively grasp the target of product development.

When the failure prediction system 1 is a CRM system for managing information about users of sellers, the information about the same type of devices or parts obtained from a plurality of users for sellers can be used crosswise. A large amount of information can be obtained and more accurate prediction can be performed. Therefore, the failure prediction system 1 is a system that can realize production information sharing, optimization, and provision of improvement policies across a plurality of companies (a plurality of users, sellers, manufacturers).

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the scope of equivalents thereof.

For example, the failure prediction system according to the present invention may evaluate the quality of welding using various information stored in the storage unit. The obtained evaluation information may be appropriately notified to the user. For example, the failure prediction system may notify the user terminal or the like in real time when a failure occurs in the device or component or the welding quality.

The configuration of the manufacturing line is an example, and the PLC, PLC-GW, and CPU for calculation can be omitted, and the physical quantity related to welding may be directly transmitted from the robot controller or the like to the network.

At least a part of the information held by the failure prediction system may be viewable from the user terminal or the manufacturer terminal. As a result, not only the seller but also the maker and the user can effectively utilize the information held by the failure prediction system.

The “distributor” is a person who sells devices or parts to the user, and when the manufacturer directly sells these to the user, the “distributor” includes the manufacturer.

Although FIG. 1 shows an example in which the respective units of the failure prediction system 1 are included in the same system, some of them may be included in different systems via the network 2. For example, the collection unit 11 and the calculation unit 13 may use SaaS different from the CRM system.

1 Failure Prediction System 2 Network 3 Manufacturing Line 4 Seller Terminal 5 Maker Terminal 6 User Terminal 11 Collection Section 12 Storage Section 13 Computing Section 14 Notification Section 15 Ordering Section 17 Sales Information Storage Section 18 Operating Information Storage Section 19 Prediction Section 20 Proposal Section 31, 31a, 31b, 31c Cell 32, 32a, 32b PLC
33 PLC-GW
34 Communication GW
35 Welding system 36 Robot main body 37 Sensors 38 Robot controller 39 Calculation CPU

Claims (7)

A sales information storage unit that is managed by a seller who sells the device or part to a user who uses the device or part, and records sales information of the device or part to the user;
A collection unit that collects operation information of the device or parts via a network,
An operation information storage unit that records the operation information of the device or the part in association with the sales information recorded in the sales information storage unit;
A failure prediction system, comprising: a prediction unit that predicts a failure time point of the device or component by machine learning based on the sales information and the operation information. The sales information is information on a sales history of the device or parts, a maintenance history of the device or parts, or a modification history of the device or parts,
The failure prediction system according to claim 1, wherein the seller is a person who sells the device or component to the user, maintains the device or component, or modifies the device or component. The sales information storage unit further records substitute product information regarding a substitute product for the device or part,
The failure prediction system according to claim 2, further comprising: a proposing unit that proposes a substitute for the device or part by machine learning based on the sales information, the operation information, and the substitute information. The suggestion unit generates a guess model for performing an evaluation when the device or part currently in use is replaced by a substitute product, and based on the evaluation of the substitute product obtained from the guess model, the product currently in use. Alternatively, the failure prediction system according to claim 3, which determines whether or not there is a preferable alternative to the component. The failure prediction system according to any one of claims 1 to 4, wherein the prediction unit predicts the failure time point based on the sales information and the operation information of the same type of the devices used by a plurality of the users. .. The failure prediction system according to any one of claims 1 to 5, wherein the failure prediction system is a system managed by the seller using cloud computing. The failure prediction system according to any one of claims 1 to 6, wherein the device is a welding or processing system, a robot body, a jig, or sensors.

Images