An image analysis system supports quality improvements and increased productivity by detecting signs of operational failures in production line facilities and deviations in worker activities on the front lines of manufacturing.

The image analysis system is expected to be able to dramatically improve the in-process guarantee rate(1) for products through a transition from “representative point management” in quality assurance based on lots to “all point management” (continuous monitoring of the status of Man, Machine, and Material) based on individual product serial numbers.

Furthermore, by shifting the role of on-site management supervisors from a focus on “after the fact” measures to the monitoring of trends and preventative measures using obtained image data, they will tie these activities into the prevention of failures before they occur.

In recent years, mega-recalls in various industries have brought about a renewed awareness of the importance of accumulating and managing manufacturing performance data to identify the causes of product defects, and to implement countermeasures.

In the advanced manufacturing workplaces of the future, it will be necessary to gather a wide range of work-related performance data, including manufacturing performance and inspection data and the results of visual checks by workers.

It aims to reach new traceability by establishing mutual links among these different forms of product performance data. All of these will be achieved by introducing new manufacturing execution systems that incorporate IoT technologies.

The image analysis system uses depth cameras to extract 3-D forms in order to measure worker activities, and obtains positional information on human joints, such as hands, elbows, and shoulders.

Then, based on frontline interviews and observations, it derives standard behaviour models that exclude information not directly related to tasks (e.g., the length of arms and legs). Furthermore, it identifies deviations in worker activities by using statistical comparisons with those standard behaviour models.

The system also detects abnormalities in the case of defects in materials and facilities by analysing differences compared to video images under normal conditions. In addition, the system is able to detect abnormalities in the case of welding defects, by combining voltage and current data from existing facilities with light emitting element colour analysis using high speed cameras.

In this way, the system quickly extracts only information related to improvements in quality and productivity from huge volumes of video data, and combines this function with data analysis to improve work efficiency and quality stabilization, and to quickly discover defects.

By accumulating image data and connecting final products based on individual product serial numbers, the image analysis system can detect production processes which cause defected products, and improve them. When inappropriate operation is detected, it can trace final products based on serial numbers, thus achieving multi-traceability.

In the future, starting at the Harima Plant, Hitachi and Daicel plan to install the image analysis system at six overseas plants, and aim to construct a globally integrated management system by aggregating and analysing information via cloud service. Hitachi will make the image analysis system available to the manufacturing industry worldwide as generalized digital solutions by applying ideas and technologies of the IoT platform “Lumada”.

(1) Non-defective product ratio in individual processes on assembly and manufacturing lines. The non-defective product ratio for final products can be increased by identifying and removing defective parts from individual processes and sending only non-defective products to downstream processes.