Since its inception, the Pulp and Paper Industry has demonstrated its ability to adapt to challenging situations and times. Today, manufacturers mostly face limited growth opportunities in mature markets. With market demand holding steady, improvements in OEE and production utilization rates have become the focus of manufacturers in both North America and Western Europe.
Though tackling these processes requires manufacturers to make numerous organizational changes, digitalization is emerging a major player in the transformation of industry. A 2019 McKinsey & Company report highlights the potential savings benefits of an automation journey, which can reduce the cost base of a producer by 15%. Of this, half of the 15% (7.5%) can be attributed to AI and Analytics. In this article, we discuss how Machine Learning and Industrial Analytics can benefit key ope rational performance metrics.
Industrial Analytics at a Glance
In the not so distant past, Machine Learning was relegated to academia, with very few real-world use cases. In the last few years Machine Learning has come into its own through commercial applications, with an upsurge in implementation throughout industries such as medicine, finance, cyber and manufacturing.
The migration of Machine Learning from the Ivory Tower to the plant floor can be attributed to a number of causes. The primary driver of change is the significant decrease in cost in data storage and transfer, as well as a drop in computational power pricing. Additionally, senior management has begun to recognize the potential impact that AI and Big Data can have in terms of revenue increase and cost savings.
A typical industrial plant contains thousands of sensors embedded in different types of machines. Although data is generated on a continuous basis, much of it is was not designed to be accessible for analysis or other operational purposes. With the exception of SCADA-based monitoring and control, most sensor data currently provides little if any value to the production facility.
SCADA monitoring is typically performed on a small number of high-priority machine assets. Control thresholds are manually determined, and in the case of a threshold breach (such as temperature, vibration, pressure etc.), technicians are alerted. Condition-based monitoring can sometimes automatically trigger additional processes.
With Machine Learning for Industrial Analytics, algorithms are applied to the huge amounts of data that plant machines generate. These algorithms are trained to identify anomalous behavior patterns and correlations of patterns. In contrast with SCADA monitoring, AI-driven Industrial Analytics analyzes all data, regardless of whether rules-based thresholds have been breached. Based on pattern recognition, the algorithm can classify deterioration, potential failure and the reasons thereof. In reliability terms, these are referred to as Time to Failure (TTF) and Root Cause Failure Analysis (RCFA).
Operations & Maintenance: A New Cost Savings Opportunity
Pulp and Paper is a decidedly automated industry that uses complicated machinery. According to the CEPI, automation has been a major productivity driver of the industry. While Paper & Board production between the periods of 1992-2018 grew by over 40% (from 65.1 Million Tons to 92.2 MT), employee numbers during this same period decreased by 56% (from 411,113 to 180,852).
Nevertheless, Operations and Maintenance in Pulp and Paper has yet to realize significant advantages from automation.
According to the Central Pulp & Paper Research Institute, maintenance costs account for 10% of sales in the industry, a substantially higher rate than the chemical industry, car manufacturing, and engine manufacturing.
Like many industries, a significant O&M cost that Pulp and Paper faces is Unscheduled Downtime. There is no specific estimation for Pulp and Paper losses due to these incidents: Pulp and Paper do not publish internal data on the cost of unplanned downtime, and at the point of writing there is little published data on the average number of production days Pulp and Paper loses per year.
Nevertheless, it is estimated that over 80% of plants across industries are unable to calculate the cost of Unscheduled Downtime. In fact, many plants underestimate costs by as much as between 200% and 300%. The explanation for this considerable number gap is simple: It is too complex to track all the cost components.
In light of this, past calculations of an hour for downtime for a single machine of $10,00 is very likely on the low-side. In fact, the losses for an entire Paper and Pulp plant from unscheduled downtime could exceed $1 million per hour. Exact calculations aside, machine downtime is extremely costly to the industry, especially when you take lost revenue opportunity cost into account.
Calculating the ROI from AI-driven Industrial Analytics
Activity-Based Costing (ABC) is the accounting term used for the methodology of accurately assigning all the direct and overhead costs incurred in a production process. The problem with ABC is that it is considered impractical and a “nice-to-have.” Activity-level data is often unavailable and it is too time-consuming to generate.
The result is that when Unscheduled Downtime is calculated, only a few of the obvious variables are considered. The purpose of this document is to outline the various cost elements included in Downtime, and to provide tools that can be used to approximate the annual Unscheduled Downtime cost.
The Opportunity Cost of Lost Production
Across all industries, the average Unscheduled Downtime for a plant is 17 days per year, though this number varies significantly both across and within industries.
In the processing industries, revenue will change based on fluctuations in commodity prices. The estimate below of the average cost for 5 different industries is a starting point, and should be used only for high-level / back-of-the-envelope calculations.
Unscheduled Downtime varies based on industry. However, even within industries, it is important to recognize variations in the incidence of Unscheduled Downtime. The following data is from a survey conducted by ARC Advisory of senior executives and engineering, operations and maintenance managers in Oil & Gas. As shown below, although the most common production loss was 3-5% per year, there is a significant range from best performers (<1%) to underperformers (>10%).
What can we learn from this data? Although industry averages for Unscheduled Downtime are interesting from a benchmarking perspective, averages must be modified to reflect the operating conditions at the plant level.
Cost Components of Unscheduled Downtime
The following costs are incurred as a result of machine failure:
- Labor Costs / Overtime: Employees who are idle until the resumption of production continue to be paid even as output and productivity come to a halt. Furthermore, overtime pay may be necessary for complex repairs or to make up for lost production.
- Labor Repair Costs: The repair costs associated with Unscheduled Downtime are higher than planned maintenance activities due to the urgency of restoring production. When Failure Root Cause data is unavailable, there is more of a reliance on trial and error, as technicians must identify the source of the failure before repairs can be made.
- Other Variable Costs: The ongoing costs of running a production line, such as utilities outlays, cannot simply be reduced because of a breakdown in a production line or even an entire plant. Overhead and other fixed expenses are still incurred regardless of whether the facility is producing at full or only partial capacity.
- Tools and Spare Parts: Spare parts and tools are used to repair broken machinery. Many production facilities invest in excess parts. A carrying cost is associated with maintaining an inventory of parts as a protection against unexpected machinery breakdowns.
- Damaged Production Output: The occurrence of a breakdown in industrial machinery is preceded by asset degradation and evolving failure. During this period, production quality is often compromised and does not meet plant standards. In these cases, inferior-quality output must be disposed of.
- Excess Inventory: To make up for the potential in lost production, plants may retain access to an inventory of finished goods as a buffer. The cost of holding inventory is typically around 10%-30% of the inventory’s value, per year.
How Unscheduled Downtime Increases Overall Maintenance Costs
The following chart depicts the costs of different asset reliability scenarios. In the case of Under Maintenance or Run-till-Failure, Unscheduled Downtime results in lost production and spikes in maintenance costs to restore production levels.
Another scenario is that plants recognize the potential cost of Unscheduled Downtime and then over-invest in both Preventive Maintenance practices. With Over-Reliability, the industrial plants perform unnecessary maintenance as a precaution. A significant cost element is planned outages – when production is stopped while parts are replaced based on a predetermined schedule and inspections are performed.
Asset Reliability Cost Scenarios
How to Calculate Unscheduled Downtime Cost in a Pulp or Paper Mill
The following calculation is used to determine Unscheduled Downtime:
The first step is to create realistic assumptions to be used in a cost calculation. In some cases, the data will be readily available or relatively easy to estimate (production capacity). For estimates of costs that are more difficult to calculate (e.g., maintenance costs), we suggest the use of industry benchmarks.
- Determine the output of a production line or plant. For this exercise, we have selected a mill with an annual capacity of 100,000 tons.
- Select a product grade. Although plants produce multiple product grades, for the purpose of simplicity, select a common product grade and assume 100% production of this grade. In this example, we have assumed a Wood Free Uncoated (WFU) product grade.
- Forecast a commodity price. Select a forecast commodity price for a specific product grade. Although pricing fluctuates, one can use a 12-month average price or analyst estimate for a 12-month outlook. For WFU, we have selected a forecast price of 800 Euros per ton.
- Estimate Current Unscheduled Downtime (days). In our example, there are 15 days per year in Unscheduled Downtime or 4.1% of lost production.
- Estimate Maintenance cost as a percentage of sales. For a mill, industry data suggest that 10% of sales are allocated to Maintenance expenses. It should be noted that if a plant is relatively new or relatively old, the allocation percentage should be decreased or increased, respectively.
In the example of this paper mill, the following calculations are used to determine the annual cost estimate for Unscheduled Downtime.
Estimate Savings from Reduction in Unscheduled Downtime.
Once we have calculated the cost of Downtime, we can calculate the cost savings from reducing Unscheduled Downtime using Industrial Analytics.
In our example, we assume that 15 days of Unscheduled Downtime costs the plant 3.5 Million Euros per year. If this is reduced to 10 days (by 33%), the annual benefit is 1.2 Million Euros. It should be noted that due to the existence of variable production raw material costs, the full 1.2 Million Euros does not equate to operating or net profit.
The calculation of Unscheduled Downtime is not simple. It requires an understanding of plant output, revenue drivers and maintenance costs. Because it is unlikely that all this data will be available, there are ways to estimate Unscheduled Downtime by modifying industry benchmark data by applying realistic assumptions.
Manually Configured Rule-Based Alerts: Why Reactive Maintenance Isn’t Enough
Pulp and Paper plants rely primarily on traditional rule-based SCADA monitoring. The downside of SCADA systems is that they are unable to potential failure in instances where the manually set thresholds were not breached. In the image below, we see an example of a SCADA monitoring system of machine temperature. If the machine’s temperature surpasses 40 degrees, an
alert is triggered. However, SCADA cannot detect abnormal sensor behavior that occurs within the threshold bandwidth.
SCADA monitoring and other types of Reactive Maintenance are the norm within the Pulp and Paper industry. Machines are “run- to-failure,” then remain dormant while repairs occur. The underlying assumption as to why plants utilize this so-called run-to-failure approach is that (a) machine failure is treated as a random event and (b) plants lack predictive tools. Reactive Maintenance is unfortunately more expensive than predictive maintenance, because it costs considerably more to repair a broken machine than it does to fix a machine while it is still operating.
One of the most compelling rationales for using Machine Learning for asset maintenance is because it detects machine failures that were previously unidentified by existing asset monitoring systems. In the example above, we see indications of anomalous sensor behavior. With Machine Learning, the algorithm will analyze this data to determine whether there is a pattern of anomalous behavior that could be indicative of machine health deterioration or potential asset failure
Using Machine Learning to detect anomalous machine behavior, reactive maintenance can be reduced and plants reap the financial benefits.
How do cost savings occur?
- Better planning means better wrench time metrics.
- Replacement parts can be ordered in advance, leading to optimized repair crew schedules.
- The pressure (and cost) of switching repair crews from routine to urgent tasks can be lessened with advanced notice.
- Machines workloads can be reduced while waiting for parts, without the need for abrupt shutdowns that bring production to a standstill.
Moreover, Machine Learning solutions that provide Root Cause Failure Analysis can provide maintenance crews with a clearer picture of why the machine is failing. As a result, O&M has less trial and error and more effective maintenance.
With Machine Learning, Paper Manufacturers also have an opportunity of reducing the cost of time-based Preventive Maintenance (PM). Time-based maintenance is founded on historic failure patterns. For example, certain assets tend to follow age-based failure patterns.
The so-called Bath-curve shown below maps the average failure rates for many electro-mechanical components and motors. The hypothesis is that during Infancy and Wear-out stage, failure rates are relatively high. With this assumption in mind, the use of Preventive Maintenance to reduce failure during the Wear-out phase is logical. Bolstering maintenance work during periods where maintenance has historically failed could prevent this failure from occurring in other assets.
This approach uses average historic data to define current maintenance schedules. However, in practice only 20% of asset failures are common and predictive – the other 80% of failures are random.
In other words, in preventive maintenance, which lacks precise knowledge of asset degradation, there is a risk and cost of over-maintenance. With Machine Learning, the decision to repair a specific piece of machine equipment is based on sensor data about the particular asset, as opposed relying on human-made decisions based on trends or educated estimations.
How Machine Learning Prolongs Asset Life Spans
Industry data indicates that machine assets in North America and parts of Western Europe tend to be older than those of their competitors in Asia and other emerging markets. For instance, in North America, the technical age of half the recovery boilers is at least 30 years old.
Machine Learning aids in the extension of machine asset lifetime because:
- Machines can be repaired before failure occurs. Repairs are based on the specific root cause of failure; there is less trial-and-error when AI-driven insights are provided to maintenance crews.
- Unnecessary over-maintenance, such as conducting Preventive Maintenance out of concern for unscheduled downtime, can be avoided.
It’s important to note that these factors do not mitigate against harmful practices such as equipment abuse, overusing equipment or not using equipment for its designed purpose.
In the past, the Pulp and Paper industry has demonstrated flexibility and resilience during challenging periods of heightened competition and external threat. Based on this impressive history, we expect the industry to leverage Machine Learning to improve its operations and finances.