You ask an Industry 4.0 AI vendor how its platform works. The vendor describes agnostic technology that takes in all of your data sources in real time and produces predictions, warnings, prescriptions and recommendations to raise yield and throughput while cutting losses. A visualization shows data flowing into a cloud and business insight coming out the other end.
For a complex process plant, that answer raises more questions than it settles.
The questions that follow the pitch
- What will you solve for us that we could not solve with our existing models?
- Much of our data is poor. How will you predict and optimize the process without understanding its fundamentals?
- Our process changes a lot and rarely runs the same way for more than a few weeks at a time. How does the platform handle that?
- Can it integrate with our existing process control and monitoring systems?
- How will our engineers and operators come to trust the platform if they do not understand the logic behind it?
- Have you done this on a facility like ours, and on which use cases?
If the vendor serves a wide spectrum of industries, you may not be satisfied with the answers. The usual next step is an offer to run its technology on your historical process data, as an offline exercise with little to lose.
The demo
If you accept, expect to spend hours, along with busy colleagues, teaching the vendor's application engineer the details of your manufacturing process. When the findings come back, they often add little. They may not go beyond what your current linear empirical models or rigorous kinetic models already tell you.
A generic vendor can be useful for simpler applications, in industries where analyzing process data and building models is not yet common and easy gains are plentiful. A plant that already runs its own process models has a higher bar, and needs a more accurate and deliberate approach to applying AI to a complex process. That leaves two options.
Building it in-house
One option is to build AI capability internally, combine it with your own process knowledge, and develop applications for specific problems. Homegrown data science groups can do this, but even the largest companies tend to avoid building their own AI products. An operating company employs its people to run plants, and it is hard for it to keep pace with a company whose whole business is the software. Most in-house products also lack a long-term roadmap, and support fades when the developers move to other roles.
A partner dedicated to your industry
The other option is a vendor that has committed itself to your industry. Solving the problems your team has not been able to solve needs current AI methods combined with deep expertise in your specific process. In practice that means:
- Engineers from your industry. The vendor's application engineers have worked for years in companies and plants like yours. They know where your most valuable and difficult problems are, what tools and models you already have, and which problems are worth going after. They speak your language, and they know how to align the right stakeholders in the plant around common objectives because they have done it before.
- Discipline with data and models. They know how to treat noisy and missing data, how to bring process first principles into the model, and how to verify that the relationships the model has learned between variables are correct.
- Experience across similar plants. Because the vendor works on plants like yours, you benefit from what it has learned applying its platform elsewhere, and from what it keeps learning.
- A structured path into the control room. Models are developed and deployed safely, through a methodical procedure that respects your process control environment. The partner works with you on the management of change (MOC) documents you need, provides training material for operators and engineers, commissions the application with you, proves its economic value, and stays to help you sustain that value for years.
It is possible to find the right solution among the many Industry 4.0 AI vendors and their ambiguous answers. The right fit is a partner dedicated to your domain that speaks the same technical language as your engineers.
How Imubit works
Imubit takes this approach in refining, chemicals and polymers, cement, and mining and minerals. Its application engineers are former process and process control engineers, and they involve plant engineers and operators at each step. The technology is Deep Learning Process Control® (DLPC®). Process engineers build the models themselves in DLPC® Builder, with no Python and no machine-learning background, and the resulting application runs in closed loop on top of the existing APC and DCS, inside limits set by plant engineers.
