Artificial intelligence is rapidly evolving. The challenge facing manufacturing companies is not so much to find good models of AI, but rather to transform AI capabilities into something usable.
Deployment of AI in industrial environment is not only about model capabilities. Data infrastructure, system integration, usability and human factors are very important as well.
AI Requires More than a Good Model
Consider an example of predictive maintenance application. Some machine learning model finds certain patterns in the data provided by the sensors, thus recognizing the equipment malfunction.
Technically, it might be considered a successful prediction.
On the other hand, the question is:
- Can the engineers access that data easily?
- Is that information understandable for them?
- Can the result of prediction be easily integrated into the maintenance process?
- Does the system give enough context for someone to take some action?
If the answers are negative, there will be little use of that information.
Human in the Loop Approach
Decisions in manufacturing processes may be based on contextual information, which cannot be contained in the machine-generated data.
Anomaly recognition algorithm may flag some unusual situation. However, the person, who knows that there were recent changes in the production process, will be able to say that it was just that. No need to take any action because of that.
Some simple workflow might look like:
Machine / Sensor Data
↓
Data Processing
↓
AI / ML Analysis
↓
Anomaly or Prediction
↓
Human Review
↓
Operational Decision
This way the scalability of AI can be combined with contextual understanding of manufacturing specialists.
Data Infrastructure Is Crucial
AI systems require high quality data. The manufacturing process may include sensors, PLCs, various machines, databases, MES platforms, and legacy systems. All of them can generate information in their own formats.
Integration of these various sources may become one of the key issues in the process of AI deployment.
Before deploying complex AI solutions, one should think about data quality, integration, security, monitoring and availability.
Even the most sophisticated model cannot compensate for poor input data.
Selecting a Good Use Case for AI
There aren’t necessarily any AI solutions to every manufacturing challenge.
Companies should begin with problems where they have access to the data and where a better prediction or analysis will deliver tangible benefits.
Some possible applications include:
- Predictive maintenance
- Quality control automation
- Optimization of production processes
- Energy consumption monitoring
- Anomaly detection
- Operation analytics
A narrower focus at the beginning may help to evaluate results and identify areas of improvement before extending the technology.
Implementation Is a Part of the Engineering Challenge
During implementation, companies should take into account who will be interacting with the system.
Engineering and operational staff must understand not only the recommendation made by the intelligent system but also the reasoning behind it and when the human decision-making process will override it. Simple and intuitive interfaces may make it much easier to integrate intelligent systems into existing processes.
Companies like PowderForge AI are exploring intelligent approaches for industrial settings and the design of more advanced manufacturing processes.
Designing Industrial AI Solutions
The future of industrial AI will not depend solely on the increasing capabilities of models. The ability to connect the model to reliable data and existing systems and workflows will also play a crucial role.
Manufacturing companies that see AI implementation as a technical and organizational challenge will have a good chance to move from experimentation to successful applications.
Building an intelligent system is one thing.
Making sure that the intelligent system is used is another.
Top comments (0)