Demand for coatings has not faltered during the pandemic. As a result, the industry remains slow in embracing digitization at scale. At the same time, new computational technologies promise to unlock the value of innovation toward higher performance coatings and streamlined processes.

But for artificial intelligence (AI) to be truly effective, large-scale data aggregation is necessary. That requires a substantial organizational shift. Let’s look at some key points and examples that reveal how you can strategically implement AI software for maximum ROI.


Why AI?

Artificial intelligence is essentially a statistical technique to optimize the things companies already do well. While employees generally possess excellent intuition and skills for delivering professional work, the computer engine takes it a step further by analyzing large datasets and uncovering patterns for further improvement.

This digital cannonball, paired with the human brain, can pierce problems more deeply and discover more unpredicted correlations than humans could alone. In that sense, AI is an extension rather than an emulation of human intelligence. For the coating industry, implementing AI can be valuable in a number of ways.


Failure Prevention

Imagine a computer that can distill patterns from a pool of input data and, with experience, enable itself to perform tasks such as:

  • Navigate a self-driving car
  • Sort bananas based on size, damage level, and ripeness
  • Predict a local thunderstorm above a crowd-filled stadium
  • Automation of stock trading
  • Challenge a doctor’s diagnosis

This is the power of machine learning technologies. In the coatings industry, it can evolve to predict failure modes like crack propagation, corrosion, creep, fatigue, color fading and diffusion of one material into another.

When fed with thousands of images of specific coatings, after a team of human experts has tagged specific failure areas, the computer can diagnose and prevent flaws in existing products with high accuracy.


Cooking Up the Perfect Formulation

Chemical discoveries are done using large-batch reactors of up to 100 liters. But digital tools like thermal imaging, infrared thermography and supervised machine learning enable these reactions to take place in microfluidic reactors containing just a few drops.

This generates less waste, ensures safety and accelerates innovation by complementing trial-and-error experimentation and mathematical simulations (in-silico synthesis).

A trained machine learning model can invent new recipes for special properties, learn to integrate additional functionalities such as UV-blocking, anti-glare, antibacterial, corrosion-resistant, conductive, flexible, heat release, anti-biofouling, hydrophobic or self-stratifying multi-coat applications.

Imagine that the algorithm detects an anomaly while looking to synthesize a new polymer. By combining historical data stored in digital labs, it may predict that this will lead to unique properties that were previously unknown.

Such a blind approach to product development may result in the discovery of breakthrough visual effects, superior impact protection for spacecraft heat shields, or self-healing properties.

Leveraging the power of AI, nothing is impossible.


Painting Robots

Based on computer vision and industrial internet of things (IIoT) analytics, AI can learn to program painting robots to follow optimal trajectories on specific parts, or exact quantities of a particular color needed for a certain effect. Lamborghini’s paint plant was the first to implement such cutting-edge technology.

Instead of painstaking manual data analysis and trial and error, AI can identify defect sources in specific parts or colors based on data from pressure regulators, metering pumps, color values, rotary atomizer turbine speeds, airflows, or joint positions and torques.

Root-cause analyses will improve the process sequence, track unknown correlations and set up predictive maintenance schedules for early intervention.

Curing ovens for car body paints undergo fluctuating conditions in terms of ambient temperature and airflow. These can lead to flaws and other anomalies that the AI will detect. Using heat-up curve simulations, it will conjure up the perfect curing schedule, further bringing the automotive industry toward its lights-out manufacturing standard.


Powder Coating

In powder coating, a dry polymeric powder is electrostatically applied with a corona gun to metal parts such as cars, bicycle frames, door frames, extrusions, building facades, fitness equipment, and industrial computer enclosures.

Even carbon composite, MDF and parts made through injection molding can be colorized and protected this way.

After coating, parts are oven-cured using heat or UV light. Because the process omits any liquid carriers, it emits fewer volatile organic compounds (VOCs), and the result is harder and thicker than paint. But let’s forget about becoming an organic chemistry tutor and remain on topic.

The powder coating industry, which includes anodizing and electro-plating, has reached $11.9 billion. To serve this steadily growing market, low-code AI platforms have been developed that guarantee process and quality stability, independent of line operator experience.

Popular options for running machine learning projects are:

Advanced predictive models can optimize curing schedules, perform fault detection, and provide a closed-loop control solution using sensor networks. AI plays a critical role in developing predictive models using algorithms such as:

  • Support-vector machine (SVM)
  • K-nearest neighbor (KNN)
  • Random forest (RF)
  • Deep learning (DL)
  • Neural networks (CNNs or ANNs)

The effect of AI-augmenting the powder coating process is that mixing and application happens more efficiently. This reduces scrap rates, cost, and environmental impact.

Service life of equipment is maximized using predictive maintenance. Novel approaches to baking result in shortened cycle times, energy savings, and perfectly smooth coatings, even with low thicknesses and uncommon substrates.


More Is More

The quality of input data to a machine learning model is paramount. For example, the accuracy of fault detection improves as scientists label damaged areas in image sets with better precision.

A human understanding of chemistry, engineering and physics is still necessary.

Yet to reach high accuracy levels, quantity is more important. Continuous input data gained from IoT devices on the shop floor as well as ongoing experimentation will generate the millions of data points needed to make a machine learning model truly insightful.

AI models, like humans, improve with time and new data. The more data, the better the AI learns.

In addition, unstructured data in internal reports, literature, data sheets and the internet become valuable assets. The challenge lies in text mining and natural language processing tools that can extract useful information, even if proprietary.

This can also prevent potential loss of institutional knowledge given the aging of the workforce.


A Full Change Agenda

As the organization becomes digitally aware with edge intelligence, changes will happen from the surface to the core. Not only does AI pertain to new product development and IT, but it also has an impact at higher levels such as:

  • Supply chain optimization
  • SKU management/inventory planning
  • Marketing programs
  • Customer service (chatbots)
  • Sales forecasting
  • Component design
  • Knowledge sharing
  • Lean production planning

In addition to enterprise resource planning (ERP) and product lifecycle management (PLM) systems, companies will be equipped with a manufacturing execution system (MES).

The MES collects failures, product quality data and process values. It reports the real-time status of equipment and sends instructions, which keeps good track of material flow and product traceability.

This way, the AI stays informed on the current condition of machinery and historical data regarding its use, empowering it to make accurate predictions for upcoming production runs.

That ultimately reduces scrap rates and raw material use, as well as optimizes quality, service life and cycle times.

When AI-readable data is collected inside all departments of the organization, a digital mindset is required. Employees need to engage in multi-disciplinary collaboration and partnerships with tech startups. Big data rather than personal opinions and office politics will lead to breakthroughs in decision making.

Staff will need to replace mistrust in AI and robots with a positive outlook toward the future and adopt new skill sets with more value-added activities — even in the face of temporary job losses.


A Bright Future

Instead of blindly throwing darts at the artificial intelligence menu, it is chiefly important to carefully investigate the added value of state-of-the-art offerings.

Sometimes the organization isn’t ready, or AI is there only to take a tiny step to improve something that already works well and provides enough customer satisfaction.

On the other hand, there are often hidden benefits or serious bottlenecks that the technology can resolve. And if it doesn’t solve any concrete problems, it at least provides more awareness about processes.

Always perform thorough risk assessment and analyze the accuracy and capabilities of new software tools and their underlying algorithmic foundations.

For any organization on the road to digital transformation, three action points should be high on the agenda:

  • Upskilling the existing workforce
  • Attracting new data talent before big tech steals it away
  • Setting up a minimum viable solution toward a connected software architecture with easy-to-use data exploitation tools