Fujitsu has announced the launch of an expert consulting service under its MONOZUKURI Total Support Solutions for Artificial Intelligence (AI) for design and manufacturing sites to support customer manufacturing. This service will be available in Japan starting October 2016. As its platform, the service will use the MONOZUKURI AI Framework to systematize functions for incorporating AI into manufacturing locations.

Low Fujitsu1 Figure 1: System structure

Low Fujitsu2 Figure 2: Predicting the number of layers of a printed circuit

MONOZUKURI AI Framework is an AI-usage platform that implements Fujitsu's AI technology, Human Centric AI Zinrai, and is made up of learning databases, AI processing engines, security authentication servers, and so forth. This platform is being offered based on the experience gained from using it as a smart manufacturing platform at Fujitsu's manufacturing locations.

The consulting service will support the implementation of AI technology in customer manufacturing locations by building a learning database for each of the variety of task processes and products in design and production locations, using the MONOZUKURI AI Framework to enable increased AI precision through continuous learning, and also by screening the collected data in response to the customer's needs and the characteristics of their products, tuning the data in order to improve prediction accuracy.

In this way, the service will contribute to the customer's manufacturing transformation, for example, by shortening design studies for printed circuit boards or enabling improved efficiency in setting up production lines, and by providing a system that continuously learns.

In order to make use of AI, it is necessary to extract knowledge and experience from the big data acquired from previous assets and locations, collect it in learning databases, and create a model with it. In manufacturing locations, however, there are all sorts of products and a variety of business processes, such as specification studies, design, validation, and manufacturing, and because standardized models which can provide the high-accuracy predictions and decision processing demanded for each of these cases do not exist, it is necessary to build appropriate models each time.

Low Fujitsu3 Figure 3: Searching for similar components in 3D models

Low Fujitsu4 Figure 4: Automatic creation of image-recognition programmes

Fujitsu has developed the MONOZUKURI AI Framework, which can distinguish between and appropriately use learning models for each business process and product, and will offer it as part of a consulting service.

Features of the MONOZUKURI AI Framework

1. Builds learning databases for each usage context

Manufacturing involves all sorts of business processes, such as specification studies, design, validation, and manufacturing, for an array of different products. In order to provide highly accurate predictions and decision processing for each of these cases, different learning databases are built for each.

2. Improved prediction accuracy through generation-management functions

By undertaking generation management of the learning databases for each cycle of product design and production, the system optimises itself through continued learning the more it is used, improving prediction accuracy. This means AI can be used even with new products under development, enabling stable operations.

3. Strong security functionality

The system protects its important learning databases through user authentication and communications encryption.

4. Easy to combine with existing systems

The system features a standardised set of web APIs, making it easy to link with customers' existing systems and programmes. This makes it possible to use AI for prediction and data generation in customers' CAD platforms and production-line machines, for example.

Example Uses of the MONOZUKURI AI

1. Design support for printed circuits in electronics design (Estimating number of circuit layers)

By using AI and inputting the features of a new product, this service can predict the number of layers necessary in the printed circuit board from the learning database. This can enable a reduction in the printed circuit board design process of approximately 20%.

2. Search 3D models for similar components in structural design

When searching for similar components with 3D models, as when searching with a 3D model of a screw, for example, this service has improved search accuracy to 96% by using AI, as compared to 68% previously. These fast and accurate searches of similar components using AI can be used in a variety of situations, such as re-using design data that used old components of similar shapes, or searching for the use of faulty components in other products.

3. Automatic creation of image-recognition programmes for production lines

With programmes that use image recognition by production line cameras to detect component positions and shapes, the correct data is learned, and it is possible to automatically create programmes that realise accurate recognition. In an internal study, compared to an image-recognition engineer developing a programme, use of this technology cut development time to one tenth, and improved image-recognition accuracy by up to 97%.

Consulting Service for Introduction of AI to Manufacturing Sites

The number of customers considering the use of AI at their manufacturing sites is increasing. However, a lack of precedents has made it difficult to actually move forward with deployment. Through its consulting service, Fujitsu will transform the manufacturing processes of its customers, by continuously improving its AI technologies through issues extraction, POC, data screening, and data tuning.