In chemical manufacturing, we must look at the process both in detail and holistically, in order to identify inefficiencies. By analyzing production disturbances through the use of process-based machine learning, we have the opportunity to reach new levels of chemical process control.


What is a “production disturbance”?

The significance of the term “production disturbance” (PD) varies since every manufacturing facility has a unique operational structure, raw materials supply, machine configuration, and production environment.

For the sake of this discussion, a production disturbance is any unintentional event in the chemical production process that leads to process inefficiencies, unplanned stoppages, rework, or scrap, for example:

  • Valve leakages
  • Head pressure drops in pumps
  • Lubricant issues, e.g. frothing
  • Inconsistent bearing temperatures

It’s important that PDs are defined specifically on the basis of individual machines, processes, and manufacturing environments.


Reducing production disturbances – The paradox of preventive maintenance

For many years, one of the most championed best practices in asset maintenance was preventive maintenance. The idea of preventive maintenance is to preempt and avoid malfunction or production disturbances by performing scheduled asset maintenance regularly.

It has since been discovered that preventive maintenance can be inefficient in a number of ways, leading to:

  • Redundant planned downtime (up to 40% of preventive maintenance costs are spent on assets with negligible failure impact.)
  • Secondary damage to equipment – caused by invasive inspections
  • Premature/untimely equipment replacement
  • Materials waste – lubricants, oils, coolants etc.
  • An inflated inventory of spare parts

These cost factors are part of what has led manufacturers to Industry 4.0, and more specifically, from preventive to predictive maintenance.


Is predictive maintenance the answer?

Predictive maintenance is without a doubt a game changer. It’s a much welcomed improvement over previous maintenance strategies, and is fast gaining recognition as the new best practice for leading maintenance operations.

Predictive maintenance is focused at preventing mechanical failure in specific assets. However, production disturbances are not necessarily asset failures.

In fact, disturbances are more often the result of process failures such as irregular cooling in a tank (the disturbing factor) that’s yielding high pressure in a pump (the disturbance), for example.

This calls for a broad-scope examination of the chemical production process and its production parameters. A narrow focus on individual asset behaviors leaves the production process context out of the equation.

Predictive maintenance is not a -one-size-fits-all solution. Not accounting for the context of the process will lead to too many false-positives and a flood of alerts that don’t provide insight and harm the credibility of the system.


And what about maximizing OEE?

Another well-known methodology for production optimization is by closely monitoring Overall Equipment Efficiency (OEE).

OEE is calculated using the formula:

Availability X Performance X Quality = OEE

The method was developed by Seiichi Nakajima in the 1960s as a means to maximize availability, performance, and quality – and in doing so, minimizing production disturbances.

OEE is a bottom-up approach that gives operators and technicians “ownership” of their assigned processes with the goal of minimizing the Six Big Losses:


The downside of the approach is that for complex facilities such as the ones in chemical processing, this formula can be too broad.

For example, the formula represents each of the components equally. This can be countered using weighted variables, but that can lead to overproduction and manipulations to the formula that don’t necessarily improve production throughput.


The solution: predict and prevent process disturbances

Focusing solely on deploying predictive maintenance or increasing the OEE percentage can lead to sub-optimization. The impact of individual sub-processes on the performance of the entire system needs to be evaluated at depth.

This leads us to the core concept behind “chemical process control”:

Using automated root cause analysis, predictive analytics, and what-if simulation to predict and prevent process disturbances that impact production throughput.


Shifting the focus to the process

While an individual pump, motor, or filter might malfunction, it is often an instability in the chemical production process – a process disturbance – which has led to the failure. In other words, the process disturbance is the root cause for the asset failure.

To tackle this complex problem, we need to account for relationships between production parameters across all stages of the manufacturing process.

The Seebo platform. A predictive alert is displayed within the context of the manufacturing process.
The Seebo platform. A predictive alert is displayed within the context of the manufacturing process.

Using process-based machine learning, we can uncover relationships that would otherwise be impossible to detect:

  1. The production plant is precisely modeled to include all the production lines, physical assets, manufacturing stages, and the product flows through the process.
  2. Production context is added through feature engineering – critical for closing the gap between the real-world manufacturing environment and data representations. Once the data is contextualized, machine learning algorithms such as Random Forest, XGBoost, Hidden Markov, and Directed Acyclic Graph can be used to form predictions. The data has been analysed by the ML algorithms within the context of the entire manufacturing process. This results in accurate predictions regarding quality levels, maintenance, and the supply chain.
  3. Personnel and management receive actionable predictions with supporting root causes in time to improve the performance of the production process.


Chemical Process Control – Optimization through AI

By using machine learning algorithms that take into account the process, we get focused and contextual predictive alerts. This is a huge opportunity for chemical process optimization since data is relatively well collected and stored in this sector already.

Leveraging this data with process-based AI means being able to pinpoint the root cause of process disturbances with extreme accuracy, and predict failures before they have the chance to affect production.

Design an Industry 4.0 solution in just two weeks

Get Demo