Hazard prediction system

Abstract

The present disclosure provides a hazard prediction system capable of performing highly accurate hazard prediction regardless of the knowledge and experience of the user. A hazard prediction system (100) comprises an information acquiring unit (101), a plan acquiring unit (102), and a prediction information acquiring unit (106). The information acquiring unit (101) acquires site information relating to a work site (WS). The plan acquiring unit (102) acquires a work plan for the work site (WS). The prediction information acquiring unit (106) acquires hazard prediction information as a result of causing a language model (LM) to interpret the site information acquired by the information acquiring unit (101) and the work plan acquired by the plan acquiring unit (102).

Description

Hazard prediction system

[0001] The present disclosure relates to a danger prediction system.

[0002] Conventionally, a program capable of favorably supporting risk prediction activities has been known (see Patent Document 1 below).

[0003] The program described in Patent Document 1 displays an interactive screen on a display unit in which input sentences and output sentences are arranged consecutively, and receives input sentences from a worker on the interactive screen that represent potential hazards at a work site. When the input sentence representing a hazard is received on the interactive screen, the program causes a computer to execute a process of displaying output sentences representing countermeasures against the hazard on the interactive screen.

[0004] Japanese Patent Application Laid-Open No. 2021-018751

[0005] However, the technology described in Patent Document 1 requires the user to consider risk factors for the work and select from multiple options or input them directly, which poses a problem in that the accuracy of risk prediction for the work depends on the knowledge and experience of the user.

[0006] The present disclosure provides a hazard prediction system that is capable of performing highly accurate hazard prediction regardless of the user's knowledge or experience.

[0007] One aspect of the present disclosure provides a hazard prediction system comprising an information acquisition unit that acquires on-site information regarding a work site, a plan acquisition unit that acquires a work plan for the work site, and a prediction information acquisition unit that acquires hazard prediction information as a result of interpreting the on-site information acquired by the information acquisition unit and the work plan acquired by the plan acquisition unit using a language model.

[0008] According to the above aspects of the present disclosure, a hazard prediction system can be provided that is capable of performing highly accurate hazard prediction regardless of the user's knowledge or experience.

[0009] Fig. 2 is a conceptual diagram showing an example of an embodiment of a risk prediction system according to the present disclosure. Fig. 3 is a configuration diagram showing an example of an information terminal constituting the risk prediction system of Fig. 1. Fig. 4 is a functional block diagram related to risk prediction by the risk prediction system of Fig. 1. Fig. 5 is a schematic diagram showing an example of a work site targeted by the risk prediction system of Fig. 1. Fig. 6 is a flow diagram showing an example of risk prediction by the risk prediction system of Fig. 1.

[0010] Hereinafter, embodiments of a risk prediction system according to the present disclosure will be described with reference to the drawings. The embodiments described below are examples and do not limit the invention. Not all features and combinations thereof in the embodiments of the present disclosure are necessarily essential to the invention. Note that identical or corresponding components in each drawing are designated by identical or corresponding reference numerals, and redundant explanations may be omitted.

[0011] Fig. 1 is a conceptual diagram showing an example of an embodiment of a risk prediction system according to the present disclosure. Fig. 2 is a configuration diagram showing an example of an information terminal IT constituting the risk prediction system 100 of Fig. 1. Fig. 3 is a functional block diagram related to risk prediction of the risk prediction system 100 of Fig. 1. Fig. 4 is a schematic diagram showing an example of a work site WS targeted by the risk prediction system 100 of Fig. 1.

[0012] The hazard prediction system 100 of this embodiment predicts potential hazards at a work site WS, such as a construction site or a building site as shown in Fig. 4, and supports hazard prediction training conducted at the work site WS. The hazard prediction training, also known as hazard prediction activities or toolbox meetings, is training in which workers engaged in work such as construction or manufacturing anticipate and point out potential hazards in their work, with the aim of preventing accidents and disasters.

[0013] Details will be described later, but the hazard prediction system 100 of this embodiment is characterized by mainly comprising an information acquisition unit 101, a plan acquisition unit 102, and a prediction information acquisition unit 106, as shown in Figure 3. The information acquisition unit 101 acquires site information related to the work site WS. The plan acquisition unit 102 acquires a work plan for the work site WS. The prediction information acquisition unit 106 acquires hazard prediction information including, for example, advice on hazard prediction training, as a result of having a language model LM interpret the site information acquired by the information acquisition unit 101 and the work plan acquired by the plan acquisition unit 102.

[0014] The configuration of the danger prediction system 100 of this embodiment will be described in detail below.

[0015] 1, the danger prediction system 100 of this embodiment is configured, for example, by one or more information terminals IT connected to a communication line NW. The danger prediction system 100 may also have, for example, one or more servers SV connected to the communication line NW.

[0016] The information terminal IT and server SV constituting the hazard prediction system 100 are connected to one or more external servers ESV via, for example, a communication line NW. The information terminal IT and server SV constituting the hazard prediction system 100 are also configured to be able to acquire, for example, the detection results of a sensor SN installed at a work site WS via the communication line NW.

[0017] The communication line NW may include, for example, a local area network (LAN) of the work site WS. The communication line NW may also include a wide area network (WAN). Examples of wide area networks include a mobile communication network terminated at a base station, a satellite communication network using a communication satellite, and the Internet. The communication line NW may also include, for example, a short-distance communication line based on a wireless communication standard such as Wi-Fi or Bluetooth (registered trademark).

[0018] The information terminal IT includes, for example, a personal digital assistant (PDA) including a smartphone or a tablet-type personal computer, a desktop personal computer (DPC), a notebook personal computer (LPC), etc. The server SV is a computer connected to the information terminal IT via a communication line NW.

[0019] The information terminal IT includes, for example, an output device OD, an input device ID, a communication interface CI, a central processing unit (CPU), an auxiliary storage device AS, and a memory device MD, which are connected by a bus, as shown in Fig. 2. The server SV may also have a configuration similar to that of the information terminal IT shown in Fig. 2.

[0020] The output device OD outputs various types of information to the user, for example. The output device OD includes, for example, a display device, a sound output device, a printer, a vibration generator, etc. The display device includes, for example, a liquid crystal display, an organic EL (Electro-Luminescence) display, an indicator lamp, etc. The sound output device includes, for example, a speaker, a buzzer, etc.

[0021] The input device ID accepts various inputs from the user. The input device ID includes, for example, an input device (mechanical input device) that accepts mechanical operation inputs from the user. The mechanical input device includes, for example, a button, a toggle, a keyboard, a mouse, a touch panel mounted on the output device OD, a touch pad provided separately from the output device OD, etc.

[0022] The input device ID may also include a voice input device capable of accepting voice input from the user. The voice input device may include, for example, a microphone capable of collecting the user's voice. The input device ID may also include a biometric input device capable of accepting biometric input from the user. The biometric input device may include, for example, a camera capable of acquiring image data containing information about the user's fingerprint or iris.

[0023] The communication interface CI is used as an interface for connecting to external devices so that they can communicate with each other. This allows the information terminal IT to communicate with external devices such as other information terminals IT, servers SV, external servers ESV, and sensors SN through the communication interface CI. The communication interface CI may have multiple types of communication interfaces depending on the communication method between the connected devices.

[0024] The CPU, for example, executes a program loaded into the memory device MD and implements various functions of the information terminal IT in accordance with instructions from the program.

[0025] The auxiliary storage device AS is a non-volatile storage device that stores the installed programs as well as necessary files, data, etc. The auxiliary storage device AS is, for example, an EEPROM (Electrically Erasable Programmable Read-Only Memory) or flash memory.

[0026] For example, when an instruction to start a program is received, the memory device MD loads the program from the auxiliary storage device AS so that the program can be read by the CPU. The memory device MD is, for example, an SRAM (Static Random Access Memory).

[0027] 1 includes, for example, an external server that provides weather information, an external server that provides a construction management system, and an external server that provides a database of dangerous cases that have occurred at work sites. The construction management system is a cloud service that streamlines the business process of construction management, including, for example, management of photographs, construction drawings, and schedules at each work site WS.

[0028] A user of the hazard prediction system 100 can obtain weather information for the work site WS at a specified date and time from, for example, an external server ESV that provides weather information, via an information terminal IT. The weather information includes, for example, information on the weather, temperature, humidity, air pressure, precipitation, solar radiation, snow accumulation, wind speed, wind direction, etc. at the work site WS.

[0029] A user of the hazard prediction system 100 can, for example, pre-register photographs of the work site WS, construction drawings, schedules, etc. in the external server ESV that provides the construction management system via the information terminal IT. Also, a user of the hazard prediction system 100 can, for example, obtain pre-registered photographs of the work site WS, construction drawings, schedules, etc. from the external server ESV that provides the construction management system via the information terminal IT.

[0030] The server SV constituting the hazard prediction system 100 may also acquire weather information for the work site WS at a predetermined date and time from an external server ESV that provides weather information and store it in a storage device. The server SV may also acquire photographs, construction drawings, schedules, etc. of the work site WS from the external server ESV that provides a construction management system and store them in a storage device. A user of the hazard prediction system 100 may also register, for example, photographs, construction drawings, schedules, etc. of the work site WS in advance in the server SV via an information terminal IT.

[0031] 4, the sensors SN include an imaging device SN1, a temperature and humidity sensor SN2, a wind direction and speed sensor SN3, a rain gauge SN4, etc., which are installed at the work site WS. The sensors SN may also include a distance measurement sensor such as a LiDAR (Light Detection and Ranging), a noise sensor, a vibration sensor, an oxygen amount sensor, a hydrogen sulfide amount sensor, a dust sensor, a rain sensor, a soil moisture sensor, etc. Furthermore, the sensors SN may also include a sensor that detects the condition of construction machinery such as a hydraulic excavator HS, and a sensor that detects the physical condition of workers working at the work site WS.

[0032] The image capturing device SN1 is, for example, a monocular camera. Furthermore, the image capturing device SN1 may be, for example, a three-dimensional camera (3D camera) capable of acquiring not only two-dimensional image information but also three-dimensional information including information on the distance to an object shown in the image and the depth of the image, such as a stereo camera, a ToF (Time of Flight) camera, or a depth camera. The image capturing device SN1 is installed at an elevated position from which the entire work site WS can be viewed, such as the roof or a support pillar of the site office SO of the work site WS.

[0033] The imaging device SN1 is connected to, for example, a communication line NW and transmits captured images to a server SV or an information terminal IT. Furthermore, the imaging device SN1 and a ranging sensor such as LiDAR, which can acquire three-dimensional information, detect the position, size, shape, etc. of objects present at the work site WS and transmit the detected information to the server SV or the information terminal IT. Objects present at the work site WS include, for example, people P such as workers, construction machinery such as hydraulic excavators HS and cranes, vehicles such as dump trucks DT, construction materials CM such as piping, safety fences SF, and cones or pylons PY.

[0034] As described above, the risk prediction system 100 of this embodiment includes the information acquisition unit 101, the plan acquisition unit 102, and the prediction information acquisition unit 106 shown in Fig. 3. In the example shown in Fig. 3, the risk prediction system 100 further includes a verbalization unit 103, a prompt generation unit 104, and a calling unit 105.

[0035] Each part of the danger prediction system 100 shown in Fig. 3 represents a function of the danger prediction system 100 that is realized, for example, by loading a program stored in an auxiliary storage device AS into a memory device MD and executing it by a CPU in an information terminal IT or a server SV. That is, each part of the danger prediction system 100 shown in Fig. 3 is realized in at least one of the information terminal IT and the server SV connected via a communication line NW.

[0036] The language model LM is implemented in the information terminal IT, the server SV, or the external server ESV depending on the capacity. The language model LM is, for example, a large language model (LLM). The language model LM is, for example, GPT-4.

[0037] The information acquisition unit 101 acquires, for example, site information related to the work site WS and outputs the acquired site information to at least one of the verbalization unit 103 and the prompt generation unit 104. The site information includes, for example, at least one of construction drawings of the work site WS, detection results of sensors SN installed at the work site WS, or weather information (web information) for the work site WS acquired via the communication line NW. In other words, the information acquisition unit 101 is configured to be able to acquire site information including at least one of construction drawings of the work site WS, detection results of sensors SN installed at the work site WS, or weather information for the work site WS.

[0038] Specifically, the information acquisition unit 101 acquires the construction drawings of the work site WS from the external server ESV by accessing the external server ESV that provides the construction management system via the communication interface CI and the communication line NW. The information acquisition unit 101 may also acquire the construction drawings of the work site WS from the information terminal IT or the server SV in which the construction drawings of the work site WS are stored in advance in the auxiliary storage device AS.

[0039] The information acquisition unit 101 may acquire the detection results of the sensors SN by, for example, directly accessing the sensors SN installed at the work site WS via the communication interface CI and the communication line NW. The information acquisition unit 101 may also acquire the detection results of the sensors SN by, for example, accessing an information terminal IT or a server SV in which the detection results of the sensors SN are stored in advance in an auxiliary storage device AS.

[0040] The information acquisition unit 101 acquires information about objects at the work site WS based on the detection results of a sensor SN, such as an imaging device SN1 or a ranging sensor. Specifically, the information acquisition unit 101 detects monitored objects from images of the imaging device SN1 by arbitrarily applying known image processing techniques such as semantic segmentation or machine learning. The monitored objects include, for example, a person P, such as a worker. The monitored objects may also include other objects present at the work site WS. Examples of the other objects include work machines such as hydraulic excavators HS and cranes, and vehicles such as dump trucks DT. The other objects may also include specific stationary objects present at the work site WS, such as construction materials CM, such as piled piping, safety fences SF, and pylons PY. The other objects may also include specific topographical features present at the work site WS, such as ditches, holes, and piles of earth and sand.

[0041] The information acquisition unit 101 also acquires site information such as the current temperature, humidity, wind direction, wind speed, and rainfall at the work site WS, detected by sensors SN such as a temperature and humidity sensor SN2, a wind direction and speed sensor SN3, and a rain gauge SN4. The information acquisition unit 101 also acquires site information such as the current noise, oxygen concentration, hydrogen sulfide concentration, dust concentration, rainfall, and soil moisture at the work site WS, detected by sensors SN such as a noise sensor, a vibration sensor, an oxygen sensor, a hydrogen sulfide sensor, a dust sensor, a rain sensor, and a soil moisture sensor. The information acquisition unit 101 also acquires site information such as the posture of the work machine, the force acting on the work machine, and the physical condition of the worker, detected by sensors SN such as a sensor that detects the condition of a construction machine such as a hydraulic excavator HS and a sensor that detects the physical condition of the worker working at the work site WS.

[0042] The information acquisition unit 101 may also acquire weather information for the work site WS on the day of work from an external server ESV, for example, by accessing the external server ESV that provides weather information via the communication interface CI and the communication line NW. The information acquisition unit 101 may also acquire weather information for the work site WS on the day of work from an information terminal IT or a server SV in which weather information for the work site WS is stored in advance in an auxiliary storage device AS.

[0043] The information acquisition unit 101 may also acquire, for example, hazardous events that have occurred in the past at other work sites. Specifically, the information acquisition unit 101 acquires hazardous events that have occurred in the past at other work sites by, for example, accessing an external server ESV in which a database of hazardous events that have occurred in the past at other work sites is stored via the communication interface CI and the communication line NW. The information acquisition unit 101 may also acquire hazardous events that have occurred in the past at other work sites from, for example, an information terminal IT or a server SV in which a database of hazardous events that have occurred in the past at other work sites is stored in advance.

[0044] The plan acquisition unit 102 acquires, for example, a work plan for the work site WS and outputs the acquired site information to at least one of the verbalization unit 103 and the prompt generation unit 104. The work plan includes, for example, at least one of text in a natural language input by a worker at the work site WS to the input device ID of the information terminal IT, or a schedule of work for the day at the work site WS.

[0045] The information terminal IT is equipped with an input device ID that can input a work plan in natural language, for example. Specifically, the input device ID of the information terminal IT includes, for example, the mechanical input device described above. In this case, a worker at the work site WS inputs a work plan for the day of work in natural language text, for example, using a desktop personal computer DPC installed in the site office SO, the keyboard of a laptop computer LPC carried by the worker, or the touch panel of a PDA. The plan acquisition unit 102 acquires the work plan in natural language text input to the mechanical input device of the information terminal IT.

[0046] The input device ID of the information terminal IT may include, for example, the aforementioned voice input device. In this case, a worker at the work site WS speaks a work plan for the day in natural language and inputs it into a desktop personal computer (DPC) installed in the site office SO or into a voice input device of a PDA or notebook computer (LPC) carried by the worker. The plan acquisition unit 102 acquires the work plan in natural language text by, for example, converting the voice input into the voice input device of the information terminal IT into text using known voice recognition technology.

[0047] The plan acquisition unit 102 is also configured to be able to acquire a work plan including a schedule for the work site WS. Specifically, the plan acquisition unit 102 accesses an external server ESV that provides a construction management system via a communication interface CI and a communication line NW, for example, to acquire the schedule for the work site WS for the day of work from the external server ESV. The plan acquisition unit 102 may also acquire the schedule for the work site WS for the day of work from an information terminal IT or a server SV in which the schedule for the work site WS has been stored in advance in an auxiliary storage device AS.

[0048] The verbalization unit 103 verbalizes, for example, in natural language, the on-site information and risk cases acquired by the information acquisition unit 101 and the schedule acquired by the plan acquisition unit 102. Specifically, the verbalization unit 103 verbalizes, for example, the information acquired by the information acquisition unit 101 by applying it to a predefined text template.

[0049] A text template representing the current weather conditions at the work site WS is specified in a format such as "The current weather at the work site is aaa, the temperature is bbb degrees, the humidity is ccc degrees, the wind is ddd, and the wind speed is eee meters," where "aaa" is the weather, "bbb" is the temperature, "ccc" is the humidity, "ddd" is the wind direction, and "eee" is the wind speed. A text template representing the location of an object to be monitored at the work site WS is specified in a format such as "There is a ggg at fff" or "There is a ggg at fff," where "fff" is the location information of the object to be monitored and "ggg" is the type or name of the object to be monitored.

[0050] It is also possible to omit the verbalization unit 103. In this case, for example, the on-site information and dangerous cases acquired by the information acquisition unit 101 and the work plan including the schedule acquired by the plan acquisition unit 102 are input to the prompt generation unit 104.

[0051] The prompt generation unit 104 generates a prompt to be input to the language model LM, for example, based on the site information acquired by the information acquisition unit 101 and the work plan acquired by the plan acquisition unit 102. The site information and work plan used by the prompt generation unit 104 to generate the prompt may be verbalized by the verbalization unit 103. The prompt generation unit 104 generates a prompt for outputting hazard prediction information corresponding to a combination of unverbalized site information or verbalized site information and an unverbalized work plan or a verbalized work plan, for example.

[0052] The prompt generation unit 104 generates, for example, a plurality of example questions and assigns them in advance to the language model LM via the calling unit 105. The example questions are defined by a combination of prerequisites (i.e., constraints), such as site information and a work plan for the work site WS, and a correct answer to be output. This allows the language model LM to understand (learn) the output format for prompt instructions.

[0053] The example questions assigned to the language model LM include, for example, site information such as two-dimensional and three-dimensional construction drawings of the work site WS, long-term and short-term weather information acquired via the communication line NW, and detection results of the sensor SN including images of the work site WS and environmental measurements. The example questions also include, for example, work plans such as various tasks that workers may perform, tasks for the day and several days later as interpreted from construction drawings and schedules, and time allocations for each task. The example questions also include, for example, correct answers such as hazard prediction information such as buried objects, numerical values, and caution points based on topography as interpreted from construction drawings.

[0054] More specifically, the site information of the example problem assigned to the language model LM includes, for example, the number of people P based on the detection results of the imaging device SN1, the condition of the ground at the work site WS, the work status of the people P, dust, the condition of the piled construction materials CM, the condition of the work site, etc. In this case, the correct answer to the example problem includes hazard prediction information related to preventing contact between people P, preventing falls, etc.

[0055] Furthermore, if the site information of the example problem assigned to the language model LM includes environmental measurements taken by a sensor SN installed at the work site WS, the answer to the example problem will include hazard prediction information related to health damage to workers. Furthermore, if the site information of the example problem assigned to the language model LM includes weather information acquired via the communication line NW, the answer to the example problem will include hazard prediction information related to slippage and deviations from work procedures caused by weather, and hazard prediction information related to health damage to workers.

[0056] The calling unit 105 calls the language model LM via, for example, a predetermined API (Application Programming Interface), inputs the prompt generated by the prompt generating unit 104 to the language model LM, and obtains the output (answer).

[0057] The prediction information acquisition unit 106 acquires hazard prediction information, which is the output of the language model LM, via the calling unit 105. Specifically, the prediction information acquisition unit 106 acquires hazard prediction information as a result of having the language model LM interpret the site information acquired by the information acquisition unit 101 and the work plan acquired by the plan acquisition unit 102. More specifically, the prediction information acquisition unit 106 acquires hazard prediction information that includes, for example, advice on hazard prediction training.

[0058] As described above, the information acquisition unit 101 can acquire risk cases corresponding to the site information and work plan of the work site WS from an external server ESV that provides a database of risk cases that have occurred at other sites. In this case, the prediction information acquisition unit 106 may acquire risk prediction information as a result of having the language model LM interpret risk cases that have occurred at other work sites, in addition to the site information and work plan of the work site WS.

[0059] The hazard prediction system 100 of this embodiment also includes an information terminal IT having an output device OD. In this case, the prediction information acquisition unit 106 generates a control signal to output the acquired hazard prediction information and outputs it to the output device OD including a display device. As a result, hazard prediction information according to the site information and work plan of the work site WS is output from the output device OD of the information terminal IT, and the hazard prediction information is displayed on the display device.

[0060] Next, the operation of the danger prediction system 100 of this embodiment will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing an example of danger prediction by the danger prediction system 100 of Fig. 1.

[0061] 5, the hazard prediction system 100 executes process P1 for acquiring on-site information. In process P1, the information acquisition unit 101 acquires on-site information of the work site WS where the worker is working from the information terminal IT, the server SV, or the external server ESV, and outputs the acquired information to the verbalization unit 103 and the prompt generation unit 104.

[0062] Next, the hazard prediction system 100 executes a process P2 for acquiring a work plan. In this process P2, the plan acquisition unit 102 acquires a work plan for the work to be performed by the worker at the work site WS from the information terminal IT, the server SV, or the external server ESV, and outputs the acquired information to the verbalization unit 103 and the prompt generation unit 104.

[0063] Next, the risk prediction system 100 executes process P3, for example, to verbalize the site information and work plan. In this process P3, the verbalization unit 103 verbalizes, for example, the site information and work plan input from the information acquisition unit 101 and the plan acquisition unit 102 in natural language and outputs the verbalized result to the prompt generation unit 104. Note that if the risk prediction system 100 does not have the verbalization unit 103 or if there is no need to verbalize the site information and work plan, process P3 can be omitted.

[0064] Next, the risk prediction system 100 executes, for example, a process P4 for generating a prompt. In this process P4, the prompt generation unit 104 generates a prompt to be assigned to the language model LM based on the site information and work plan input from the information acquisition unit 101 and the plan acquisition unit 102, and the verbalized site information and work plan input from the verbalization unit 103. The prompt generation unit 104 outputs the generated prompt to, for example, the calling unit 105.

[0065] Next, the danger prediction system 100 executes a process P5 for calling the language model LM, for example. In this process P5, the calling unit 105 calls the language model LM and assigns the prompt generated by the prompt generating unit 104 in the previous process P4 to the language model LM.

[0066] Next, the risk prediction system 100 executes a process P6 to acquire risk prediction information. In this process P6, the prediction information acquisition unit 106 acquires, via the calling unit 105, risk prediction information that is, for example, the response of the language model LM to the prompt given in the previous process P5.

[0067] Next, the risk prediction system 100 executes, for example, process P7 to output risk prediction information. In this process P7, the prediction information acquisition unit 106 generates a control signal to output the risk prediction information acquired in the previous process P6 to the output device OD of the information terminal IT, and outputs it to the output device OD. After that, the risk prediction system 100 ends the processing flow shown in FIG. 5.

[0068] The danger prediction system 100 of this embodiment can be used in the following situations by executing the processing flow shown in FIG.

[0069] The first scenario in which the hazard prediction system 100 is expected to be used is when a worker uses it on-site before each task, such as excavation work or loading work at a work site WS.

[0070] Specifically, the worker inputs the details of the work to be performed at the work site WS in natural language into the hazard prediction system 100 via the touch panel or voice input device of the PDA he or she carries. The worker can also input additional information about the work, such as that the work is being done in a hurry, that the work will be done in cooperation with a work machine such as a hydraulic excavator HS, or that the worker is a beginner. The plan acquisition unit 102 of the hazard prediction system 100 then acquires the input work details and additional information as a work plan and outputs the acquired work plan to the verbalization unit 103 and the prompt generation unit 104.

[0071] The information acquisition unit 101 of the hazard prediction system 100 acquires site information including, for example, environmental measurement results detected by a sensor SN installed at the work site WS where workers perform work, construction drawings of the work site WS, and weather information for the work site WS via a communication line NW. The information acquisition unit 101 outputs the acquired site information to the verbalization unit 103 and the prompt generation unit 104.

[0072] The prompt generation unit 104 generates a prompt based on the site information and work plan input from the information acquisition unit 101 and the plan acquisition unit 102, and the verbalized site information and work plan input from the verbalization unit 103, and outputs the generated prompt to the calling unit 105. The calling unit 105 calls the language model LM and assigns a prompt to the language model LM. The prediction information acquisition unit 106 acquires hazard prediction information obtained as a response from the language model LM via the calling unit 105, and outputs a control signal for outputting the acquired hazard prediction information to the display device or audio output device of the PDA.

[0073] As a result, for example, hazard prediction information including indications of dangerous locations at the work site WS and indications of trap-prone situations is displayed on the display device of the PDA, and voice related to the hazard prediction information is output from the voice output device of the PDA. The hazard prediction system 100 can also communicate with workers via the input device ID of the information terminal IT, such as the voice input device or touch panel of the PDA.

[0074] This prevents danger to workers due to deviations from the initial plan at the start of work, and reduces worker impatience. Specifically, it avoids dangers such as workers taking shortcuts on walking routes and workers approaching heavy machinery such as hydraulic excavators HS before the machinery has completely stopped. Therefore, it is possible to reduce the burden on the construction manager of the work site WS to patrol the site to check safety.

[0075] A second situation in which the danger prediction system 100 is expected to be used is, for example, a situation in which danger is to be notified to workers working at a work site WS.

[0076] In this case, the information acquisition unit 101 of the hazard prediction system 100 acquires, for example, environmental measurement results of the work site WS detected by the sensor SN, measurement results of the worker's physical condition, and site information including weather information provided by the external server ESV. The information acquisition unit 101 also acquires hazardous cases that have occurred at other work sites from, for example, the information terminal IT, the server SV, or the external server ESV. The plan acquisition unit 102 also acquires work plans, such as a schedule for the work site WS where the worker will be working, from, for example, the information terminal IT, the server SV, or the external server ESV.

[0077] As a result, the prediction information acquisition unit 106 can acquire hazard prediction information including, for example, reports of situations that could lead to accidents and reports of workers' poor physical condition by having the language model LM interpret the worksite information, hazard cases, and work plans. Furthermore, the prediction information acquisition unit 106 notifies the acquired hazard prediction information to workers via the output device OD of the information terminal IT. This allows the hazard prediction system 100 to notify predicted hazards at the worksite WS and provide hazard prediction training and work assistance to inexperienced workers.

[0078] A third scenario in which the risk prediction system 100 is expected to be used is, for example, when providing advice regarding safety during the construction phase when creating design drawings or construction drawings.

[0079] In this case, the person in charge of creating construction drawings, etc., inputs information necessary for construction, such as the purpose and intent of the construction, the planned number of workers, timing, and work plan, into the hazard prediction system 100, for example, via the input device ID of the information terminal IT. The information acquisition unit 101 of the hazard prediction system 100 acquires, as site information, the information entered by the person in charge, the construction drawings, meteorological information and ground information provided by the external server ESV, etc. The information acquisition unit 101 also acquires hazard cases that have occurred at other work sites, for example, from the information terminal IT, the server SV, or the external server ESV. The plan acquisition unit 102 also acquires, for example, the work plan entered by the person in charge.

[0080] As a result, the prediction information acquisition unit 106 can acquire hazard prediction information including, for example, past accident cases and indications of danger due to work time pressure, by having the language model LM interpret the site information, hazard cases, and work plans. The prediction information acquisition unit 106 outputs the acquired hazard prediction information to the output device OD of the information terminal IT. This reduces the frequency of dangerous situations during construction at the work site WS. The hazard prediction system 100 can also be used when creating construction drawings and when using the drawings.

[0081] As described above, the risk prediction system 100 according to this embodiment includes an information acquisition unit 101, a plan acquisition unit 102, and a prediction information acquisition unit 106. The information acquisition unit 101 acquires site information related to the work site WS. The plan acquisition unit 102 acquires a work plan for the work site WS. The prediction information acquisition unit 106 acquires risk prediction information as a result of having the language model LM interpret the site information acquired by the information acquisition unit 101 and the work plan acquired by the plan acquisition unit 102.

[0082] With this configuration, the hazard prediction system 100 of this embodiment can perform highly accurate hazard prediction regardless of the knowledge and experience of users such as workers and designers at the work site WS. In other words, the hazard prediction system 100 of this embodiment can perform highly accurate hazard prediction according to each individual work site WS by providing the language model LM with prompts based on site information and work plans specific to the work site WS, utilizing the vast knowledge of the language model LM.

[0083] The hazard prediction system 100 of this embodiment also includes an information terminal IT having an output device OD that outputs hazard prediction information. With this configuration, the hazard prediction system 100 can notify users such as workers and designers of the hazard prediction information via the output device OD of the information terminal IT.

[0084] In addition, in the danger prediction system 100 of this embodiment, the information terminal IT has an input device ID that can input a work plan in natural language. With this configuration, a user of the danger prediction system 100 can input a work plan to the danger prediction system 100 by speaking in natural language or by text input via the input device ID of the information terminal IT.

[0085] Furthermore, in the hazard prediction system 100 of this embodiment, the plan acquisition unit 102 is configured to be able to acquire a work plan including a schedule for the work site WS. With this configuration, a user of the hazard prediction system 100 can check hazard prediction information corresponding to the work of the day before starting work at the work site WS without having to input a work plan at the work site WS.

[0086] Furthermore, in the risk prediction system 100 of this embodiment, the information acquisition unit 101 is configured to be able to acquire site information including construction drawings of the work site WS.

[0087] With this configuration, the hazard prediction system 100 of this embodiment can acquire hazard prediction information based on construction drawings by using the language model LM to interpret site information, including construction drawings. Specifically, for example, if the construction drawings include information about buried objects, a worker inputs a command in natural language via the input device ID of the information terminal IT to output key points about the work to be performed at the work site WS. As a result, the worker can acquire hazard prediction information via the output device OD of the information terminal IT, including advice such as, "Areas where buried objects are present will be marked with buried object marking tape. If tape is present, manually excavate to check for buried objects," or "A large amount of excavated soil is expected to accumulate on the construction road when work is completed. Be careful of the hydraulic excavator's parking position."

[0088] Furthermore, in the hazard prediction system 100 of this embodiment, the information acquisition unit 101 is configured to be able to acquire on-site information including the detection results of the sensor SN installed at the work site WS. With this configuration, the hazard prediction system 100 of this embodiment can acquire hazard prediction information corresponding to real-time on-site information of the work site WS including the detection results of the sensor SN.

[0089] Furthermore, in the hazard prediction system 100 of this embodiment, the information acquisition unit 101 is configured to be able to acquire site information including weather information for the work site WS. With this configuration, the hazard prediction system 100 of this embodiment can acquire hazard prediction information corresponding to site information specific to the work site WS, including weather information for the work site WS.

[0090] Furthermore, in the hazard prediction system 100 of this embodiment, the prediction information acquisition unit 106 acquires hazard prediction information including advice on hazard prediction training. With this configuration, even inexperienced workers can perform highly accurate hazard prediction training in accordance with the site information and work plan of the work site WS. Furthermore, since advice on hazard prediction training in accordance with the site information and work plan specific to the work site WS can be acquired, even experienced workers can perform hazard prediction training without feeling bored.

[0091] Furthermore, in the hazard prediction system 100 of this embodiment, the prediction information acquisition unit 106 acquires hazard prediction information as a result of having the language model LM interpret hazard cases that have occurred at other work sites in addition to site information and work plans. With this configuration, the user of the hazard prediction system 100 can acquire hazard prediction information based on site information of the work site WS where work is performed and hazard cases that have occurred at other work sites that are similar to the work plan.

[0092] As described above, according to this embodiment, it is possible to provide a danger prediction system 100 that is capable of performing highly accurate danger prediction regardless of the user's knowledge or experience.

[0093] The preferred embodiments of the present disclosure have been described above. However, the invention according to the present disclosure is not limited to the above-described embodiments. Various modifications, substitutions, etc. may be applied to the above-described embodiments without departing from the scope of the invention according to the present disclosure. Furthermore, each of the features described with reference to the above-described embodiments may be combined as appropriate as long as there is no technical contradiction.

[0094] This application claims priority based on Japanese Patent Application No. 2024-073737, filed on April 30, 2024, the entire contents of which are incorporated herein by reference.

[0095] 100 Hazard prediction system 101 Information acquisition unit 102 Plan acquisition unit 106 Prediction information acquisition unit ID Input device IT Information terminal LM Language model OD Output device (display device) SN Sensor WS Work site

Claims

1. A hazard prediction system comprising: an information acquisition unit that acquires on-site information related to a work site; a plan acquisition unit that acquires a work plan for the work site; and a prediction information acquisition unit that acquires hazard prediction information as a result of interpreting the on-site information acquired by the information acquisition unit and the work plan acquired by the plan acquisition unit using a language model.

2. The danger prediction system according to claim 1, comprising an information terminal having an output device that outputs the danger prediction information.

3. The danger prediction system according to claim 2, wherein the information terminal has an input device that can input the work plan in natural language.

4. The risk prediction system according to claim 1, wherein the plan acquisition unit is configured to be able to acquire the work plan including a schedule for the work site.

5. The risk prediction system according to claim 1, wherein the information acquisition unit is configured to be able to acquire the site information including construction drawings of the work site.

6. The hazard prediction system according to claim 1, wherein the information acquisition unit is configured to be able to acquire the on-site information including the detection results of sensors installed at the work site.

7. The risk prediction system according to claim 1, wherein the information acquisition unit is configured to be able to acquire the site information including weather information for the work site.

8. The danger prediction system according to claim 1, wherein the prediction information acquisition unit acquires the danger prediction information including advice on danger prediction training.

9. The hazard prediction system described in claim 1, wherein the prediction information acquisition unit acquires the hazard prediction information as a result of having the language model interpret the on-site information, the work plan, and also hazardous incidents that have occurred at other work sites.

Images