モニター

HACARUS is working with Tokyo Electron, a leading manufacturer of semiconductor manufacturing equipment, on an Al tool with the goal of reducing occupational injuries at semiconductor manufacturing plants.​

Always-on, powered by Camera x Edge AI

Real-time detection of hazards, incidents, and near-misses.

The manufacturing site is constantly monitored with a camera, which Hacarus' AI analyzes, automatically detecting incidents, such as signs of hazards and dangers in real time, alerting workers and operators as they occur.

UI

AI Usage examples in Cleanroom environments

  • Enables continuous monitoring, which is difficult for human operators
  • Immediately alerts workers and safety managers
  • Automatic recoding of incidents, including prior and following the alert, allow for analysis and future prevention activities to be performed.

Detects unsafe conditions

Detects unsafe conditions

AI detects inappropriate use of safety bars and goggles

Monitor the actions of workers

Monitor the actions of workers

AI detects worker who are unsure how to operate equipment and provides alert

System Outline

Customer

Select the area

Select the area you want to monitor for safety

Decide the scenario

Decide the scenario and actions to be monitored

HACARUS

network configuration

STEP1

Select the following to configure network configuration
・Number of network cameras
・Perform inference process for AI
・Number of edge PCs
・Host PC

Retrain AI models

STEP2

・Retrain AI models as needed - all locally; no data leaves deployment site
・The AI model is updated by capturing video of the situation to be monitored on site

start deployment

STEP3

Adjustments and testing, then start deployment & usage

   

Advantages of introduction

POINT1

Real-Time Detection

Able to detect hazards in real-time, which greatly improves the ability to prevent incidents before they occur

POINT2

Continuous Monitoring

Able to provide constant monitoring, which is difficult to do with safety measures dependent on human supervision. In addition, the image feed before and after an alert is automatically recorded, making it possible to check and review the circumstances afterwards.

POINT3

Resource Efficiency

The tool is highly resource efficient, able to train and re-train AI models from small data. This leads to cost reduction not only at the time of introduction, but also after the start of operation when models are updated.

Target Customer​

Semiconductor manufacturing, as well as industries that use cleanrooms in their manufacturing process.​

Case Study

Collaboration with Tokyo Electron

Tokyo Electron is committed to extensive safety management efforts in semiconductor manufacturing clean rooms. The company aims to reduce work related hazards focusing on efficiency and automation of safety management operations to achieve this goal - and strives to create in a more secure working environment.

Safety thinking

Claen RoomThe Tokyo Electron Group, based on the concept that "all people place the highest priority on safety and health and strive for active and continuous improvement in safety and health promotion," has established a safety management system, and are implementing safety management initiatives such as hazard prediction and safety training.

Manufacturing technology has evolved so far that even a minor mishap or a brief stoppage of a line can led to a major business loss, and thus safety management also needs to evolve.

Utilization of AIxCamera

CameraIn a large-scale manufacturing site such as a semiconductor plant, many workers are engaged in a variety of tasks. It is not realistic from the standpoint of manpower planning and cost to have human workers constantly monitoring all work sites. However, incidents and occupational injuries occur when there is no supervision.

Prior solutions used AIxCamera mainly for security purposes, but intrusion detection modules weren't sufficient for safety use cases such as:

1. Ensuring that workers are wearing protective equipment properly
2. Detection of hidden dangers at the worksite
3. Working behavior monitoring

There were also a requirements regarding the environment in which the solution would operate. Typical, cloud-based monitoring solutions do not meet high security and detection speed requirements. In addition, the solution needed to be deployed onsite with real time detection capabilities.

Deployment steps

Start by selecting the area to be monitored, then the situations and actions to be detected. Based on these inputs variables, the system design can be determined, including the number of network cameras, the number of edge PCs for AI inference, the host PCs, and the network connecting them. In parallel, AI model is enhanced as necessary. Video of the situations and actions to be monitored is taken onsite to update the AI model. There are use cases where the system can be used without AI model enhancement, so re-training will be determined based on use case details. individual. This re-training process can also be performed onsite, with no data leaving the customer site in the process.

Step

Once the equipment has been installed and the AI model updated for the customer environment, it can be put into operation after final adjustments and testing. When the AI detects an incident, it notifies workers directly with LED lights and also notifies the responsible person at a remote location via a host PC, who can review the situation from the recorded video.

Value Proposition

The newly developed AI tool not only help turn safety management tasks from analogue to digital, but also thanks its digital nature, allows for new tasks to be performed that were previously impossible such as:

Monitor

1. Instant Incident Detection
As the entire process, from image capture to AI analysis is completed on-site, it is possible to catch incidents in real-time as events unfold - this greatly increases the possibility of preventing incidents from turning into mishaps as they occur.

2. Perpetual Monitoring
The tool enables constant monitoring, which is difficult to do with safety measures dependent on human supervision. In addition, the image feed before and after an alert is automatically recorded, making it possible to check and review the circumstances which led to an incident, enabling enactment of future countermeasures.

3. Cost Efficiency
The tool is highly resource efficient, enabled by HACARUS' expertise in sparse modeling, and experience in developing AI models from small data, overcoming the limitations adherent to conventional AI, which typically require large amount of data for training and retraining (in this case image data). This leads to cost reduction not only at the time of introduction, but also after the start of operation when models are enhanced.

Future Outlook

Tokyo Electron is committed to promoting safety management through DX. In cooperation with many partners, the company will continue its efforts to achieve a safer and more secure working environment.

Subscribe to our newsletter

Click here to sign up