This browser is not actively supported anymore. For the best passle experience, we strongly recommend you upgrade your browser.
| 4 minutes read

Maintenance required: How new digital tools can solve age-old manufacturing problems

Artificial intelligence and machine learning have emerged as ubiquitous terms in global business. In many manufacturing circles, however, adopting next-generation digital tools has often been slow, insufficient, and disjointed. This is a missed opportunity, especially in plant environments where maintenance issues, equipment performance, legacy sensors and decision control systems (with inherent risk), and efficiency programs are all under the microscope. 

We’ve helped clients use specific AI and machine learning applications to implement critical changes over the span of mere months, resulting in significant equipment reliability gains, and tens of millions of dollars in profit improvement. 

Understanding the complementary roles of data, AI, and machine learning

To understand why these results aren’t more widely enjoyed, we need to ask why AI and machine learning are often sidelined. A key reason, among many,  is the fact that humans are reliable problem solvers. Give a team of motivated experts a chance to address issues that negatively impact profitability and, more often than not, they’ll figure out a fix. The primary problem: These solutions are often time-consuming, costly, and distracting. 

The fact is, AI and machine learning are excellent ways to speed up problem-solving and more effectively protect the bottom line. Fortunately, it doesn’t take a costly and complex conversion to a full-blown smart factory or highly automated system to achieve meaningful impact. Our experience shows that factories can be made smarter by implementing a series of specific steps rather than an expensive overhaul. The key is to embrace digital tools as an extension of the traditional workforce, and a way to rapidly improve your existing playbook.

While this applies across the entire manufacturing landscape, AlixPartners has found it to be particularly helpful in capital intensive energy and process industries. Chemical manufacturing, mining and metals, power and utilities, and oil and gas offshore structures (particularly the scale deepwater assets) and refineries all possess a trove of data – much of which is severely underutilized. Data, when properly collected, analyzed, and deployed, can reduce unplanned downtime, enhance production, and improve profitability.

Leave the firefighting behind

To demonstrate what is possible, consider the work AlixPartners did for a global steel manufacturer with a problem that mirrors the challenges many manufacturers face in this area.

The company was operating a steel mill with relatively new machinery. Nevertheless, the mill was experiencing thousands of small equipment failures annually, leading to a frequent production stops and profit losses.

To understand the problem and design a focused solution that would improve operations, we combined a traditional, maintenance-based reliability improvement program with a digital approach to diagnosing and resolving what was causing the equipment failures. Maintenance crews were constantly in “firefighting mode,” responding to the latest breakdown, often at the expense of preventive maintenance plans. As hard as they worked, they could not overcome a problems caused by underinvested sustaining capital during the last steel downturn. Additionally, constant and accelerating turnover among the operations and maintenance crew led to a vicious cycle of recurrent plant failures. 

To break out of this cycle, a more customized solution was needed. Building it, required a comprehensive, unemotional and fact-based understanding of what was driving the equipment failures.

This is where advanced digital tools came in handy. Coupling AI and machine learning with deep root cause analysis empowered reliability engineers and data scientists to collaborate on a solution. In this case, AlixPartners ran predictive regression analyses in Python using all available plant data—temperature, pressure, weight measurements, visual sensors, etc.—to understand the variables that were most correlated with the failures. We then developed AI predictive analytics algorithms to determine which specific operations and maintenance variables could be used to address different causes of the failures. The functional interplay between operations, maintenance, reliability engineering, safety and integrity management and digital science, with new information, provided a new safe space to re-evaluate entrenched beliefs.    

Steps to success

Building these AI-based predictive algorithms quickly requires a strategic approach to collecting, processing, and segmenting the incoming plant data to find those variables that are most effective at predicting the probability of a failure mechanism occurring. How does that happen?

  • Identify the specific failure mechanisms that represent where most of the value is being lost. Address these costliest events first, building models that quantify how increases or decreases in different variables are associated with an increase or decrease in, for example, the average time to failure. 
  • Coordinate with manufacturing personnel to rapidly create an initial list of data elements that could be influencing the observed failures. Sophisticated data cleaning and aggregation exercises deliver a set of input variables that inform the predictive model.
  • Deploy advanced machine learning algorithms like XGBoost to determine which variables are the most important predictors of failure. 
  • Simulate how different temperatures, pressures, weight measures, and other sensor values influence the likelihood of failure and the likelihood of a costly increase in time between failure and repair.   

These insights led to tangible results. In the steel mill, they allowed us to identify operating thresholds that can be established for key operating variables, which reduced equipment failures. In one example, the optimum water temperature threshold was established for water-cooled panels lining the inside of a conveyor system connecting car, in order to reduce the chance of a failure. 

A graph of water temperature

Description automatically generated

In another example, the cumulative weight of scrap metal loaded onto a conveyor belt that might cause an equipment failure was determined, so that plant personnel could adjust operating parameters going forward, to stay under that limit. 

A graph with different colored lines

Description automatically generated

In this example, AlixPartners supported a metals processing company in combining a digital approach with reliability improvements achieved through traditional root cause analysis. This led to significant equipment reliability improvement. Within six months the plant had turned the corner; equipment failures declined by more than 40%, and annual profits improved by tens of millions of dollars.

By gathering and analyzing plant data, and creating tailored operations and maintenance improvement programs, our team can apply this approach across energy and process industries to increase production, improve equipment reliability, and enhance operations and profit outcomes. Please reach out to get the conversation going. 

To read more on our perspectives on Mining click here.

Tags

article, energy, americas, english us