Hazard information extraction system and hazard information extraction model

Abstract

To provide a risk information extraction system and a risk information extraction model that can extract risk information highly relevant to a work word.SOLUTION: A risk information extraction system 100 includes an acquisition unit 110 and a risk information extraction unit 112. The acquisition unit 110 acquires a work word related to work scheduled to be performed. The risk information extraction unit 112 inputs the work word acquired by the acquisition unit 110 to a risk information extraction model, and outputs risk information highly related to the work word and ranked in risk degree to a predetermined presentation unit 114. The risk information extraction model is a model learned by machine learning using work-risk correspondence information for learning as learning data, and ranks risk information listed in a risk information database 108 and outputs the risk information.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a danger information extraction system and a danger information extraction model. [Background technology]

[0002] There is a system that searches for and provides information on past disasters, accidents, etc.

[0003] For example, there is a safety information provision system that provides appropriate safety information in response to input (see Patent Document 1). This safety information provision system can provide disaster information on past disaster cases by searching using various information, such as identifying sets of related words that are mutually related from learning information by machine learning. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-121210 A Summary of the Invention [Problem to be solved by the invention]

[0005] In such technical fields, systems for preventing industrial accidents are required. For example, in chemical plants, information on many industrial accident cases and near misses (hereinafter sometimes abbreviated to "near miss") has been accumulated and used. However, in risk prediction activities carried out before actual work, it was common to extract and use past cases based on the user's knowledge and experience. When extracting cases based on the user's experience in this way, it is expected that the appropriate cases that should be extracted may not be extracted.

[0006] The object of the present disclosure has been made in consideration of the above-mentioned problems, and aims to provide a danger information extraction system and a trained model that can extract danger information that is highly relevant to work words. [Means for solving the problem]

[0007] The danger information extraction system disclosed herein is a danger information extraction system including an acquisition unit and a danger information extraction unit, wherein the acquisition unit acquires task words related to task to be performed, and the danger information extraction unit inputs the task words acquired by the acquisition unit into a danger information extraction model and outputs danger information that is highly related to the task words and has been assigned a danger level ranking to a specified presentation unit.In this danger information extraction system, the danger information extraction model is a model trained by machine learning using learning task-danger level correspondence information as learning data, and assigns a danger level ranking to danger information recorded in a danger information database and outputs the danger information.

[0008] In addition, in the danger information extraction system disclosed herein, the learning task-risk level correspondence information may include task information and risk level information corresponding to each piece of task information, the task information may include information on dangerous behavior or conditions, and the risk level information may include information on the magnitude of the risk.

[0009] In addition, in the risk information extraction system of the present disclosure, the risk information extraction unit may output, with respect to the risk information, at least one of a correlation value between the work word and the risk information and the number of cases of the risk information.

[0010] In addition, the danger information extraction system of the present disclosure may further include the presentation unit, and the presentation unit may have at least one of a sorting function by specified items and a narrowing down function in response to user operation for the display format of the danger information.

[0011] In addition, the danger information extraction system disclosed herein may include at least one of company-wide industrial accident information and department-specific near-miss information as the types of danger information recorded in the danger information database, and may perform the danger ranking according to each type.

[0012] In addition, the danger information extraction system disclosed herein may accept a user evaluation of the danger information, and adjust the output of the danger information of the danger information extraction model or the presentation of the danger information by the presentation unit based on the accepted user evaluation.

[0013] In addition, the danger information extraction system of the present disclosure may be configured to re-learn the danger information extraction model through machine learning using the learning data in which the user evaluation is reflected in the learning task-risk level response information.

[0014] In addition, the danger information extraction system of the present disclosure may exclude danger information that includes predetermined specific words from the output or presentation of the danger information.

[0015] The danger information extraction model disclosed herein is a danger information extraction model trained by machine learning using learning task-risk level correspondence information as learning data, in which task words related to the acquired task to be performed are input and danger information that is highly related to the task words and has a risk level ranking is output. Effect of the Invention

[0016] According to the risk information extraction system of the present disclosure, it is possible to obtain the effect of being able to extract risk information that is highly relevant to work words. [Brief description of the drawings]

[0017] [Figure 1] FIG. 13 is a diagram showing an example of a risk assessment result according to an embodiment of the present disclosure. [Diagram 2] 1 is a schematic block diagram illustrating a hardware configuration of a danger information extraction system according to an embodiment of the present disclosure. [Diagram 3] 1 is a block diagram illustrating an example of a functional configuration of a danger information extraction system according to an embodiment of the present disclosure. [Figure 4] FIG. 13 is a diagram showing an example of a display screen on which danger information is displayed. [Diagram 5]FIG. 2 is a diagram showing a process flow of the danger information extraction system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] Hereinafter, embodiments of the technology of the present disclosure will be described with reference to the drawings.

[0019] First, an overview of the risk information extraction system proposed in this disclosure will be described.

[0020] First, the risk information extraction system uses the risk assessment results of each department as training data and learns the extent of damage (risk level) as a risk information extraction model using techniques such as machine learning. The risk assessment results are an accumulation of correspondence between the work content and risk level for various work tasks. The risk assessment results are an example of the work-risk level correspondence information for learning of the present disclosure.

[0021] The risk assessment result includes work information and hazard level information. An example of a risk assessment result according to an embodiment of the present disclosure is shown in FIG. 1. As shown in FIG. 1, the risk assessment result includes, for example, a classification for each task, a hazard level, a work location, and a work content. Note that the risk assessment result does not matter whether the task is an actual experience of an actual worker or an assumption. The classification, work location, and work content are examples of work information of the present disclosure, and the hazard level is an example of hazard level information of the present disclosure.

[0022] Examples of classifications include work classifications for work and types of near misses. Work classifications are classifications of the type of work, such as practical work or clerical work, and near miss types are the types of near misses for that work. Information on the level of danger is ranked according to the degree of danger, with A being the most dangerous, and D being the most dangerous. The work location is the place where the work is performed. The work content is a specific description of the work in the situation in which the near miss occurred, and can be said to be information on dangerous actions or conditions. Note that the elements used in the classifications and risk levels are merely examples, and other items and elements may be used. For example, the level of damage, expressed numerically, may also be used as the level of danger and combined with the rank.

[0023] Using the risk assessment results described above, a risk information extraction model is trained, and previously submitted or accumulated near-miss information and industrial accident information are automatically ranked and extracted according to risk. The extracted results can be displayed in order of risk (in descending order of damage). Risk rankings can also be manually revised or adjusted.

[0024] Secondly, the risk information extraction system clearly indicates the relationship between equipment, work content and risk in the extraction results. For example, near misses and types of industrial accidents can be linked to equipment and locations, enabling classification and analysis.

[0025] Thirdly, the hazard information extraction system can use different dictionaries for extraction, for company-wide industrial accident information and for near-miss information by department, to expand related words and find similar cases over a wider range. For example, it can search for chemical hazards or flammable substances in other factories as similar items.

[0026] Fourth, the risk information extraction system makes it possible to extract cases from which unnecessary data has been removed by excluding files containing specific words or setting words not to be searched.

[0027] The hardware configuration of the risk information extraction system 100 is shown in Fig. 2. It has a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.

[0028] The CPU 11 is a central processing unit, and executes various programs and controls each part. That is, the CPU 11 reads a program from the ROM 12 or the storage 14, and executes the program using the RAM 13 as a working area. The CPU 11 controls each of the above components and performs various arithmetic processing according to the program stored in the ROM 12 or the storage 14. In this embodiment, an observation evaluation program is stored in the ROM 12 or the storage 14.

[0029] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores various programs including an operating system, and various data.

[0030] The input unit 15 includes a pointing device such as a mouse, and a keyboard or a microphone for voice input, and is used for various inputs.

[0031] The display unit 16 is, for example, a liquid crystal display, and displays various information. The display unit 16 may function as the input unit 15 by adopting a touch panel system.

[0032] The communication interface 17 is an interface for communicating with other devices such as terminals. For this communication, for example, a wired communication standard such as Ethernet (registered trademark) or FDDI, or a wireless communication standard such as 4G, 5G, or Wi-Fi (registered trademark) is used.

[0033] Next, the functional configuration of the risk information extraction system 100 will be described.

[0034] Fig. 3 is a block diagram showing an example of a functional configuration of the danger information extraction system 100 of this embodiment. As shown in Fig. 3, the danger information extraction system 100 includes an information storage unit 102, a learning unit 104, a model storage unit 106, a danger information database 108, an acquisition unit 110, a danger information extraction unit 112, and a presentation unit 114. The presentation unit 114 may be provided outside the danger information extraction system 100 and may be linked to the outside.

[0035] The risk assessment results are stored in the information storage unit 102. The risk assessment results are used to train a risk information extraction model.

[0036] The learning unit 104 performs machine learning using the risk assessment results for learning stored in the information storage unit 102, and stores the learned hazard information extraction model in the model storage unit 106. The hazard information extraction model takes work information as input for the learning data, ranks the hazard level for hazard information listed in the hazard information database 108, and learns parameters so as to output hazard information that is the correct answer. Any deep learning method may be used for learning the parameters. The hazard information extraction model performs learning separately for company-wide industrial accident information, which is a dictionary of the hazard information database 108, and near-miss information by department.

[0037] The model storage unit 106 stores the risk information extraction model learned by the learning unit 104. The risk information extraction model is trained so that, when an activity word is input, the model refers to the risk information database 108, assigns a risk ranking to the described risk information, and outputs the risk information.

[0038] The danger information database 108 stores danger information to be referenced during extraction. The danger information is divided into dictionaries for company-wide industrial accident information and near-miss information by department. By dividing the dictionaries into whole company and department in this way, it is possible to extract danger information limited to a specific environment.

[0039] In addition to company-wide industrial accident information and near-miss information by department, the risk information database 108 can also include external industrial accident information, such as a collection of industrial accident cases outside the company. The risk information can also include information on the weather on the working day, the worker's years of experience, the worker's physical build, etc. The degree of risk may be corrected using this supplementary information.

[0040] The acquisition unit 110 acquires task words related to the task to be performed. The task words acquired are those input by the user. The task words are, for example, a plurality of words or sentences related to the task, such as "cleaning the reactor circulation pump strainer."

[0041] The danger information extraction unit 112 inputs the work words acquired by the acquisition unit 110 into the danger information extraction model, and outputs danger information that is highly related to the work words and has a ranked degree of danger. In addition, the danger information extraction unit 112 calculates and outputs a correlation value between the work words and the danger information, and the number of cases of danger information, for the data of danger information extracted by the danger information extraction model. The correlation value may be learned and output by the danger information extraction model, or may be calculated from the output result of the danger information.

[0042] The presentation unit 114 presents the risk information output by the risk information extraction unit 112 in a predetermined display format. The presentation unit 114 also has a sorting function and a filtering function for each item in response to a user operation. The presentation unit 114 also presents risk information that excludes specific words. For example, since risk information such as "minutes" and "list of horizontal deployment cases" is often not what the worker is looking for in the extraction, it is expected that risk information that includes these specific words will be excluded.

[0043] FIG. 4 shows an example of a display screen 50 displaying danger information (the display screen 50 is displayed on the display unit 16). Display on the display screen 50 is performed by the function of the presentation unit 114. As shown in FIG. 4, the display screen 50 displays search results 56 of danger information for the work word entered in the search box 52. In addition, the item narrowing function 54 can be used to narrow down the items for searching. Examples of items include causes and events. The search results 56 display the danger information in a list, and a details screen of the danger information can be displayed by clicking a details button. The list can be sorted in ascending or descending order by the magnitude of damage as an item. In addition, as statistical information 58, information indicating a correlation value between the work word and the danger information, and the number of cases of danger information is displayed. In addition, a check box or the like may be provided in the statistical information 58 so that the search results 56 can be narrowed down. Note that the danger information displayed in the search results 56 is data referenced from the danger information database 108. With such search results 56 of danger information and statistical information 58, the user can consider countermeasures in advance, assuming possible accidents.

[0044] In addition, the display screen 50 may be configured to accept a user evaluation for the display of the danger information. When the evaluation is accepted, an evaluation button indicating good / bad, etc. may be provided. For example, the output or display of the danger information may be adjusted using the user evaluation. When adjusting the output of the danger information of the danger information extraction model, the learning unit 104 may weight the parameters of the danger information extraction model and re-learn them. When adjusting the display, the presentation unit 114 may determine a priority order for the presentation of the danger information, and adjust the priority order according to the user evaluation. In addition, the display screen 50 is an example of a display format of the present disclosure.

[0045] Next, the operation of the risk information extraction system 100 will be described with reference to Fig. 5. The CPU 11 reads out a program from the ROM 12 or storage 14, expands it in the RAM 13, and executes it, whereby the CPU 11 functions as each unit of the risk information extraction system 100 and performs the following processes. It is assumed that the risk information extraction model has been learned in advance by the learning unit 104.

[0046] In step S100, the acquisition unit 110 acquires task words related to the task to be performed. In addition to the task words, the acquisition unit 110 accepts designation of company-wide industrial accident information or near-miss information by department. Good too.

[0047] In step S102, the danger information extraction unit 112 inputs the work words acquired by the acquisition unit 110 to the danger information extraction model, and outputs danger information that is highly related to the work words and has a ranked degree of danger. 。

[0048] In step S104, the presentation unit 114 presents the danger information output by the danger information extraction unit 112 by displaying it on the display screen 50, for example.

[0049] In step S106, the presentation unit 114 determines whether or not a user operation has been received. The user operation here includes a sorting function by each item, a narrowing down function, other screen transitions, etc. If a user operation has been received, the process proceeds to step S108, and if a user operation has not been received (for example, if the presented display screen 50 has been closed), the process ends.

[0050] In step S108, the presentation unit 114 executes a function according to the user operation to re-present the danger information. For example, when the user executes a narrowing down function, the display unit 114 narrows down the danger information to the specified one and displays it on the display screen 50. When the user executes a sorting function, the display unit 114 displays the danger information on the display screen 50 according to the specified sorting conditions. In addition, the display unit 114 displays the danger information on the display screen 50 according to the operation.

[0051] With the configuration and operation described above, the risk information extraction system 100 according to this embodiment can extract risk information that is highly relevant to work words.

[0052] The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the spirit and scope of the present invention.

[0053] Furthermore, although the present specification has been described as an embodiment in which the program is pre-installed, the program may also be provided by being stored on a computer-readable recording medium. [Explanation of symbols]

[0054] 100 Danger Information Extraction System 102 Information storage section 104 Learning Department 106 Model Memory Section 108 Danger Information Database 110 Acquisition Department 112 Danger information extraction unit 114 Presentation section

Claims

1. A danger information extraction system including an acquisition unit, a danger information extraction unit, a danger information extraction model, and a danger information database, The acquisition unit acquires an operation word related to an operation to be performed, The danger information extraction unit inputs the work word acquired by the acquisition unit into a danger information extraction model, and outputs danger information that is highly related to the work word and has a ranked degree of danger to a predetermined presentation unit. In the risk information extraction system, The task-risk correspondence information for learning used to learn the risk information extraction model is learning data in which task information is associated with risk information corresponding to each of the task information for each pair of task classification and near-miss type, The work information includes information on dangerous actions or conditions, and the risk level information includes information on the magnitude of the risk, The danger information extraction model uses the learning task-danger level response information as learning data, and through machine learning, assigns danger levels to the danger information recorded in the danger information database, and its parameters are trained to output the correct danger information.

2. The risk information extraction system according to claim 1 , wherein the risk information extraction unit outputs, with respect to the risk information, at least one of a correlation value between the work word and the risk information and the number of cases of the risk information.

3. The display unit further includes:

3. The danger information extraction system according to claim 1, wherein the presentation unit has at least one of a sorting function and a narrowing-down function for the display format of the danger information according to a user operation, the sorting function being based on a predetermined item.

4. The type of danger information recorded in the danger information database includes at least one of company-wide industrial accident information and department-specific near-miss information, and the danger ranking is performed according to each type.

5. A danger information extraction system as described in any one of claims 1 to 4, which accepts a user evaluation of the danger information, and adjusts the output of the danger information of the danger information extraction model or the presentation of the danger information by the presentation unit based on the accepted user evaluation.

6. A risk information extraction system as described in claim 5, wherein the risk information extraction model is re-learned by machine learning using the learning data in which the user evaluation is reflected in the risk response information for learning work.

7. The danger information extraction system according to any one of claims 1 to 6, wherein danger information containing a predetermined specific word is excluded from the output or presentation of the danger information.

Images