predictive maintenance using machine learning

The ideas behind predictive maintenance have been around for decades. For today, it needs to be retooled with concurrent technology – in particular what A.I. would do for predictive maintenance. In fact, almost 20 years ago, in a book called “Condition-based maintenance and Machine Diagnostics” by Williams, Davies and Drake (1994) they came out with formal definitions and the need for predictive maintenance. They defined condition monitoring simply as:

“The continuous or periodic measurement and interpretation of data to indicate the condition of an item to determine the need for maintenance”

Background

Predictive maintenance or Condition-based maintenance (CBM), focuses on identifying failures before they occur. CBM incorporates inspections of equipment at predetermined intervals to determine system condition. These inspections could be nothing more than continual data collection from the equipment about the vibrations (sound), temperature, pressure, light, voltage, current, field strength and so on. Depending on the outcome of a continual inspection, either a preventive or no maintenance activity is performed. CBM may employ several fault or defect detection methods to make fault prognostications. In general most of the methods work by comparing current sensor or inspection data with standardized reference data. 

A.I. in predictive maintenance
The A.I. based prognostics layer is the heart of the predictive maintenance system

Today with the advent of machine learning and big data technologies, CBM has suddenly shot into the limelight once again. A recent report by Global Industry Analysts states that, “With industrial consumers becoming increasingly concerned about rising maintenance costs and production downtime due to unscheduled breakdown of machinery, predictive maintenance solutions such as machine condition monitoring is gaining traction.” The report goes on to identify the largest and fastest growing industry niches for this application of predictive analytics. Vibration monitoring equipment has the largest market share and interestingly thermography equipment (infrared scanners and imagers) is stated as the fastest growing segment in this report. Finally, the report estimates that the condition monitoring market is expected to exceed $2B in less than 5 years. 

The value from CBM is well recognized by the big players in the industry. In 2009, Boeing and GE Aviation jointly developed standards for condition based monitoring stating that it “provides a 10-fold increase in real time performance of the Open System Architecture for Condition Based Maintenance (OSA-CBM) standard, making it practical for embedded health monitoring of aircraft systems”. Another report claimed that adopting predictive maintenance can reduce maintenance budgets by 30-40%!

Given that in manufacturing today, there are nearly 300,000 or so small to mid-sized companies, who can also benefit from predictive maintenance, how can this solution be made available to them? As mentioned in this article “These organizations can also benefit from having preventative strategies, which can keep maintenance and repair in-house and at a smaller cost, as well as defray initial operating costs for new and growing operations. Many issues shared by larger enterprises, like machine breakdowns, unplanned downtime, low productivity and the costs of replacing machinery, are exacerbated for smaller businesses, which need to stretch each of their resourced dollars even further”. 

Sustainable Plant recently published a report on predictive maintenance, where equipment suppliers also provide CBM technology that works with their components and monitor real time data using the cloud. This allows smaller sized manufacturers to take get the benefits of predictive maintenance with minimal additional investment. 

The role of A.I. in predictive maintenance

The underlying architecture of the CBM process is fairly uniform, irrespective of the end applications. The Machinery Information Management Open Systems Alliance  (MIMOSA) organization provides the open systems architecture for CBM, termed OSA-CBM which includes essential elements of such a system and communication protocols between the elements.

The hub and spoke model simply shows that the different components may reside on different IT platforms but from a process point of view, there are seven separate layers which can be imagined sequentially as well:

  1. data acquisition
  2. data manipulation (transformation of raw data for data mining and machine learning models)
  3. condition monitoring (provide simple alerts based on operating limits)
  4. health assessment (generate diagnostic records based on trend analysis, if health is degraded)
  5. prognostics (generate predictions of failure using machine learning models, estimate future life)
  6. decision support (recommendations of next actions), and finally
  7. human interface layer (provide access to the information in an easily digestible form)

The heart of the CBM system is layer 5 (prognostics) which is basically artificial intelligence (A.I.) or more precisely machine learning that uses component degradation and failure data from layer 3 (condition monitoring) and trend analysis data monitored in layer 4 (health assessment) to provide predictions about component failures. So to answer the headline question: what A.I. stands for in predictive maintenance is pretty much the heart of the system!

Predictive maintenance can utilize the advances in technology and the affordability which predictive analytics has brought to the market, and thus is going to be more relevant now than ever before.

Originally posted on Wed, Jan 22, 2014 @ 07:05 AM

No responses yet

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

simafore.ai