Manufacturing AI Agent

Yokohama Rubber and California-based dotData have begun jointly developing an AI agent intended to help engineers form hypotheses, interpret experimental findings and make decisions in manufacturing research and development.

The two companies announced the project in mid-October. Before applying it to live R&D work, they will verify which functions the agent actually needs and assess its technical feasibility, rather than deploying at scale first and refining afterwards.

How the System Is Meant to Work

The agent combines three inputs. The first is internal technical documentation, including design information, experimental reports and research reports. The second is the accumulated knowledge of the engineers themselves. The third is features extracted from measurement data using dotData’s Signal Intelligence automated feature engineering technology.

That third element is the interesting one. Rather than testing only the variables engineers have already flagged, the system explores patterns across measurement and experimental data without being restricted to pre-identified factors. The resulting quantitative features, read together with design information and domain knowledge, can surface relationships that existing hypotheses or individual experience would not reach.

The agent is explicitly not built to produce a single correct answer. It operates on a human-in-the-loop basis, presenting multiple possibilities, commentary from different technical perspectives and potential risks, across materials, design, analysis, production and evaluation.

Engineers Keep the Decision

Responsibility for interpretation and decision-making stays with the engineering team. Engineers form hypotheses from what the agent presents, test them through experiment or analysis, and feed the results back into the next round of discussion with the system. The companies are also examining how engineers can communicate their analytical objectives and areas of interest to the agent, including the settings required for automated feature engineering.

The intent is to broaden the range of possibilities an engineer considers and to support professional development through sustained collaboration with the tool, rather than to compress the development cycle by removing judgement from it.

Built on an Existing Framework

The project extends HAICoLab, the proprietary AI utilisation framework Yokohama Rubber has promoted since 2020. The name stands for humans and AI collaborating, and the framework has already been applied to systems that predict rubber physical properties and tyre characteristics, and to compound design against specified property targets. Yokohama had previously used dotData’s feature discovery in work on mixing process quality stabilisation.

Yokohama Rubber plans to apply the new agent to real manufacturing R&D themes and to develop HAICoLab practice further. dotData, for its part, says it will continue combining the quantitative context produced by automated feature engineering with AI agents, with a view to serving a wider range of manufacturing R&D environments.