February 2018

Trends and Resources

Business Trends: The impact of industrial analytics on the chemical industry

The opportunities for industrial analytics within the chemical industry, and how advances in machine learning can impact revenue, operations, employee safety and capital investments, are analyzed here.

Vesely, E., Presenso

The opportunities for industrial analytics within the chemical industry, and how advances in machine learning can impact revenue, operations, employee safety and capital investments, are analyzed here.

In both operational and financial measurements, the chemical industry has outperformed other sectors in the last 15 yr. Balancing this optimism, a March 2017 report by management consultancy McKinsey & Co. suggested that “the golden era may be coming to an end,” and pointed to slowdowns in total returns to shareholders (TRS) and return on investment capital (ROIC). McKinsey forecasts decelerated growth over the next decade and variations by region. As can be seen in FIG. 1, although overall growth is expected to remain at 3.6%, it is anticipated that Europe and Japan will underperform as most new growth comes from emerging markets in Asia. To what extent can Industry 4.0 contribute to margin improvement in this uncertain environment?

Fig. 1. Chemical industry forecast growth rate, 2017–2019. Source: BASF.
Fig. 1. Chemical industry forecast growth rate, 2017–2019. Source: BASF.

While multiple aspects of digitalization and Industry 4.0 are applicable to the chemical industry (e.g., 3D printing, robotics and augmented reality), this article is limited to the topic of industrial analytics for predictive asset maintenance. Specifically, the following four topics will be covered in detail:

  1. Production output
  2. Operations and maintenance (O&M)
  3. Occupational health and safety (OHS)
  4. Plant asset life.

Is the chemical industry ready?

In a 2016 survey of 222 chemical industry executives, consultancy PwC demonstrated a progressive approach to digitalization. Three results stand out: companies expect to re-invest 5% of revenues in digitalization in the next 5 yr; 75% of respondents forecast a “high level” of digitalization in the next 5 yr; and the most germane finding from this study is that in 5 yr, 88% of respondents indicate that a high significance will be placed on the gathering, analysis and utilization of data for decision making.

Taken together, the PwC report indicates a commitment on the part of senior management to embrace industrial analytics. This is an important first step.

Background: Industrial analytics for predictive asset maintenance

It was not long ago that machine learning was primarily a topic of studies within the confines of academia. However, the last few years have seen a surge in commercial applications, especially in financial services and the industrial domain.

What has changed? Significant reductions in the cost of storage and transfer of big data are matched with ever-increasing computational power. Due to advances in the cloud, artificial intelligence applications are more easily accessible and affordable. Studies by PwC and many other organizations indicate a top-down effort on the part of senior management to use Industry 4.0 as a strategic lever for improved growth and productivity.

Industrial plants have thousands of sensors embedded in machine equipment. Most of the data generated from these sensors is not captured or used in a meaningful way. The exception is the supervisory control and data acquisition (SCADA)-based monitoring of a small number of high-priority sensors (e.g., temperature or vibrations) that are used to track the health of a machine.

The traditional SCADA approach monitors whether human-set control thresholds have been breached. For instance, if a machine temperature exceeds the set limit of 40°C, this overheating will trigger further actions. With machine learning, an algorithm is trained to detect abnormal data patterns and correlations between patterns, regardless of whether or not the control thresholds have been breached. Advanced artificial intelligence enables the measurement of time to failure (TTF) and the application of root cause analysis (RCA).

Machine learning contains multiple methodologies for failure detection. Perhaps the best known is the digital twin, which relies on physical modeling of the machine and involves mechanical engineers and data scientists working together to build the representing machine model. This is one form of supervised machine learning, where the algorithm is “trained” on the underlying production asset by using physical equations, data labels or classification examples provided by humans. Once it is trained, the algorithm can apply the classification to new data.

Another option that is gaining traction is the so-called unsupervised machine learning methodology, where human expert knowledge is not needed to generate the machine digital model. Instead, vast amounts of data are analyzed, and the algorithm itself generates a model based on statistical features detected and extracted from the data.

The differences between supervised and unsupervised machine learning are not merely technical. Supervised machine learning is more resource-intensive, which impacts the ROI on industrial analytics solutions based on this methodology.

Improving production output

A major challenge for chemical plants is the cost associated with unscheduled machine downtime. Since companies do not share shutdown information publicly, it is often challenging to calculate asset failure at an industry-wide level. Nevertheless, according to Accenture’s ICIS Chemical Business, the missed profit opportunity alone on a major cracker shutdown in the US Gulf Coast was $1.4 MM/d per world-scale cracker. Furthermore, a calculation by the Aberdeen Group indicates that 2%–5% of production is lost in the petrochemical sector.

It should be noted that the chemical industry has achieved productivity gains over the past few years, thereby raising the bar for incremental gains.

Industrial analytics provide chemical plants with two important pieces of information. First, when an alert is generated with an accurate TTF, plant maintenance staff can schedule repairs in a way that minimizes disruption to production. Secondly, root cause failure analysis helps limit the likelihood of the failure reoccurring elsewhere in the plant.

A simple but powerful equation is driving the adoption of industrial analytics: lower machine downtime leads to higher yield rates and increased revenue.

Lower O&M budgets

Fig. 2. Industrial analytics shifts resources away from reactive maintenance activities. Source: Presenso.
Fig. 2. Industrial analytics shifts resources away from reactive maintenance activities. Source: Presenso.

Reducing O&M budgets is an oft-stated goal of many plants. Until recently, this goal was difficult to achieve. Industrial analytics reduces O&M spending by shifting resources away from reactive maintenance activities (FIG. 2).

The closer the repair activity to the repair incident, the greater the cost of maintenance. Advanced planning improves “wrench time,” an important O&M metric. Asset failure that is detected close to occurrence can lead to a disruption of production and of other repair activities. Tight timelines are a source of pressure on the crew. Without sufficient information about the root cause of failure, trial and error may be necessary. When machine repair is reactive, the likelihood of inefficiencies is increased. Time is wasted waiting for spare parts to arrive or for additional resources to be brought onsite. The combination of TTF and RCA are drivers for better planned repair activities, and therefore lower O&M budgets.

A sometimes-overlooked factor that also impacts O&M budgets is the relatively high cost of spare parts for the chemical industry. Practices such as maintaining unnecessary spare parts, and performing redundant and unnecessary preventive maintenance, can also be limited when a more accurate and timely view of machine degradation and asset failure is available.

Improved OHS

Fig. 3. Non-fatal injury and illness rate per 100 employees, 2016. Source: BLS measure of incidences per 100 employees.
Fig. 3. Non-fatal injury and illness rate per 100 employees, 2016. Source: BLS measure of incidences per 100 employees.

Given the nature of the industry, OHS standards are stringent. In the US, evidence can be seen of compliance with high standards. The US Bureau of Labor Statistics (BLS) publishes annual reports of illness and fatality rates across various industries, and the chemical industry has one of the best performance records, as shown in FIG. 3.

Unfortunately, some significant tragedies with recorded fatalities have occurred, such as the Bhopal Toxic Gas Leak, which was partially attributed to poor maintenance practices. The shadow of these tragedies acts as a reminder of the consequence of maintenance errors.

The use of industrial analytics improves occupational safety in a number of ways. The primary way is intuitive: the shift away from reactive maintenance results in fewer accidents.

Another driver of improved safety standards is that the learning algorithms replace human inspections. By definition, the very act of sending reliability workers to physically check machinery increases the likelihood of an incident. In the same way that robotics reduce the number of accidents by factory workers, the use of machine learning reduces the need for extraneous maintenance.

Extending plant asset life

The average age of chemical plants is significantly higher in North America and Europe relative to Asia and the Middle East (FIG. 4).

Fig. 4. Average age of ethylene cracker, yr. Source: Accenture.
Fig. 4. Average age of ethylene cracker, yr. Source: Accenture.

The well-known Bathtub Curve graph in FIG. 5 depicts the rise in failure rate as machinery ages. As noted, aging equipment is a contributing factor to the increasing frequency of unplanned stoppages. At present, many plant owners choose to over-invest in maintenance by using expensive, time-based preventive maintenance. The alternative of unscheduled downtime is both expensive and disruptive to operations.

Fig. 5. Hypothetical asset failure rate vs. time, Bathtub Curve. Source: Presenso.
Fig. 5. Hypothetical asset failure rate vs. time, Bathtub Curve. Source: Presenso.

With advances in industrial analytics, chemical plants can receive early alerts of evolving failures, providing the opportunity to remediate before the degradation leads to a shutdown. In addition, root cause failure analysis maintenance crews can focus on the underlying reasons for failure. As a result, the life of a chemical plant asset can be extended.

Of course, industrial analytics is only one element of an overall program to extend asset life. Plants may also need to reverse deterioration by performing heavy maintenance repairs and by using equipment as designed.

Challenges ahead

Numerous factors will determine when the benefits of industrial analytics are fully realized. For example, many chemical plants are not capturing and storing the sensor data that is generated, which is the first step toward the implementation of machine learning-based solutions. HP

The Author

Related Articles

From the Archive

Comments