Data Analytics for Industrial Applications: Part I, A History of Predictive Maintenance
Maintenance philosophies can be classified into two categories, these are reactive and proactive maintenance. As the name indicates, reactive maintenance is unplanned, it is simply taking corrective actions as an equipment exhibits fault symptoms that humans can sense or it just breaks down.
Proactive maintenance can also be broken down into two categories: 1- Preventive maintenance (PM) 2- Predictive maintenance (PdM). Both are planned operations to help reduce the operation and support costs and increase equipment life.
According to the data of Statista in the manufacturing sector:
“As of 2020, 76% of the respondents reported following a proactive maintenance strategy, while 56% used reactive maintenance (run-to-failure).”
According to the Market Research Future:
“The global predictive maintenance market is expected to expand at 25.5% CAGR to reach USD 23 billion in 2025 during the forecast period.”
History of Maintenance Philosophies
From the historical perspective of maintenance, one can say that the most spectacular changes have occurred in the last sixty years following World War. Until then, corrective maintenance was the only option for a maintainer where equipment used to be fixed or replaced on a breakdown basis. Nevertheless, corrective maintenance is still in use for simple components such as light bulbs or a basic pipeline which are less risky and where the failure consequences are not fatal.
From the 1950’s, mechanisation and automation steps have risen due to the increasing intolerance of downtime and the significantly increasing cost of labour. Improved machinery was of lighter construction and ran at higher speeds provoking wear out more quickly which led to the development of proactive maintenance.
Preventive maintenance is a sub-discipline of proactive maintenance in which the maintenance tasks are performed periodically. Periods are fixed intervals determined by using historical data (e.g. MTBF: Mean Time Between Failures) and without any input from the actual equipment being used. Equipment is serviced on a routine schedule whether the service is actually needed or not. However, both reactive and blindly proactive (preventative maintenance) maintenance approaches have financial and safety implications associated with them. Routine inspection rounds and lubrication, bi-monthly bearing replacements, or maintenance inspections and overhauls on aircraft systems are some of the examples of preventative maintenance activities.
In the late 1970s, the effectiveness of conducting preventative maintenance started to be questioned. A common concern about ‘over-maintaining’ arose which led to the development of predictive maintenance. Adaptively determined scheduling of maintenance actions are the main features of predictive maintenance that distinguish it from preventive maintenance. On the contrary, predictive maintenance is limited to those applications where the cost and consequences are critical and technically feasible.
From the 1980’s systems became progressively more complex in nature, bringing a more competitive marketplace and intolerance of increased downtimes. For instance, daily loss of revenue due to downtime is £320,000 for a Boeing 747 aircraft. Increasingly, risk analysis and environmental safety issues have become paramount. New concepts such as condition monitoring and expert systems have emerged. The Institute of Asset Management was established in the UK in the mid-90’s which has received significant attention from most organizations.
Since 2000’s, terms such as prognostics and industrial internet (Industry 4.0) have emerged. Today’s sophisticated sensor technology enables us to track degradation processes and empower for prognostic reasoning of equipment being monitored. With the help of advancements in computing power and sensing technology and the adaptation of artificial intelligence, aforementioned processes seem to be automated more and more.
Predictive Maintenance
As per a study 82% companies have faced unplanned downtime and the lost revenue cost runs into millions of dollars. McKinsey and Company study concludes that AI based predictive maintenance gives rise to a 10% reduction in annual maintenance costs, 25% reduction in downtime along with a 25% reduction in inspection costs.
Predictive maintenance practices have significant advantages in reducing the support and operating costs and leading to a more effective planning and operational decision making. An unexpected one-day stoppage in the machinery industry may cost up to $225,000.
Predictive maintenance helps equipment operators perform maintenance tasks when the need arises. The necessity concept is determined by assessing the health condition of the equipment continuously.
The process starts with acquisition of data and transmitting it to the higher level where the signal is processed (e.g. feature extraction and selection) which is then followed by the practice called diagnostics to detect and isolate faults (e.g. FMECA failure mode analysis). Based on the determined health condition of the system, a prognosis can then be made by employing time series trending algorithms which help estimate remaining useful life of the asset.
Potential-functional failure curve (P-F Curve) is a known way of illustrating the health condition of a system. Performance of the system is represented with this curve as seen below. As expected, it declines over time leading to a failure at the end. The aim is to detect the fault before it reaches the critical levels and to allow maintenance personnel to get prepared and supply parts for the maintenance.
Performing maintenance preparation when the system is up and running has a great effect on reducing the operation and support costs. In addition to the reduced down time, the inventory cost will be reduced as more time will be available for obtaining required parts. Moreover, the efficiency in logistics & supply chain will be increased by means of better preparation for maintenance. Eventually, the life cycle cost of the equipment will be reduced, as they are used until the end of their lives.
Until now we have talked about the types of maintenance strategies and benefits, as well as the history behind it. Next post will be about how Tarentum AI helps their customer perform predictive maintenance effectively by automatically detecting faults early.
