Developing and implementing AI tools for automation

AI tools for automation

Implementing artificial intelligence (AI) in a business or product is complex and requires considerable understand and investment. However, the rewards are also likely to be significant too. Research from Global market Insights suggests AI in manufacturing will reach €14.2 billion turnover by 2025, driven by IIOT.

AI technology and edge computing make products and processes smarter, and life more secure, intuitive and convenient. They use compact deep-learning algorithms and instrumental knowledge for efficient reinforcement learning and efficient time-series big data analysis. Furthermore, they will impact across a company’s value chain. Business needs to consider how AI  can help them, and the best place to deploy it.

But where does that leave smaller companies who identify opportunities but do not have access to the core technology? One answer comes from companies like Mitsubishi Electric who are making available some of their original AI technology.

AI tools for automation

Now under its new “Maisart” brand, the company’s proprietary AI technology is being made available to help accelerate AI-based equipment businesses and promote the wider applicability of AI in diverse business fields, including high-level information processing. Implementation of compact AI in FPGAs is one of the early products.

Deep learning can perform high-level inference, but computational needs can be costly and requires significant memory because of deep learning’s multi-layer network structure. Optimising its circuit architecture Mitsubishi Electric improved its core technology’s efficiency for implementation in FPGAs. The resulting solution can reduce inference-computational time to just one-tenth that of conventional AI.

Expanding the scope of AI applications

Real-time inference can be performed in embedded small-scale FPGAs, as well as embedded CPUs. Implementation in FPGAs helps lower hardware costs for real-time processing in applications requiring AI, such as high-precision mapping and lowers power consumption. Compact hardware AI will support the expansion of AI application to including home appliances, elevators, high-precision maps and factory automation.

AI in edge computing software

Edge computing provides real-time data analysis and enables users to create diagnostics rule by conducting offline analysis of shop floor data and then executing real time diagnostics of production systems during operation. It improves the accuracy of detecting equipment anomalies during real-time diagnostics using waveform recognition technology to learn and recognise data such as sense or waveform patterns. Finally, it implements prevent preventative maintenance and quality improvements using statistical diagnostic tool and very multivariable regression analysis

AI learning to control equipment

This technology uses model-based artificial intelligence to control equipment autonomously. It constructs models of the equipment through repeated trial and error and then learns control rules based on them. In a demonstration using a circular maze a ball is driven to its centre by tipping and tilting the maze. The technology successfully learnt how to drive the ball to the goal without the need for human programming. This machine learning based control technology will reduce the cost and time needed to develop control programs in the future.

Force feedback

To support increased robot assembly, Mitsubishi has developed a fast force-feedback control algorithm for industrial robots using AI. In tests, the algorithm shortened assembly insertion times by about 65 percent without requiring the robots to move violently. The algorithm improves the efficiency of robotic tasks for assembling electric components, PCBs and connecters, and inserting mechanical parts. The AI control makes fast, precise changes to the velocity and other parameters. It supports the incorporation of high-precision force-sensor data without stopping, as with conventional robots.

In summary, the adoption of AI in manufacturing organisations driven by the Internet of things, cost pressures and availability of skilled labour. The above examples of AI tools for automation represent the tip of the iceberg.