Integrating Advanced Analytics into the existing predictive Maintenance plan, will reinforce industrial companies’ proactive approach by using A.I models to predict failures at a very early stage. So under Industry 4.0, anomaly prediction using machine-learning based software is now a complete part of Predictive Maintenance. Proven solutions do exist in the market, which can be applied to all industries.
Let’s first adopt a common understanding about Predictive Maintenance definition.
I- Maintenance types definitions and conventions
There are many types of Maintenance performed by industries and there is usually a difference in the classification of those types and their exact scope depending on different “schools”. Typical types encountered could be Corrective Maintenance, Preventive maintenance, Conditions Based Maintenance, Predictive Maintenance, condition-monitoring…
There may be sometimes confusion between Condition-Based Maintenance and Predictive maintenance since they are often encompassed under Preventive Maintenance which could then be either (NF standard) systematic (calendar based) or Condition based or predictive.
The purpose here is not debate over the types of maintenance but just for the sake of a common understanding, let’s adopt the following simple classification:
Corrective Maintenance: Reactive approach once significant signs appear (Alarms, noise, high vibration…)
Preventive Maintenance: Based on calendar or Operating hours, and combined with some parameters thresholds (material wear, clearance limits, …)
Predictive maintenance: This is the proactive approach based on a set of data collection and analysis, such as vibration spectrum, detailed oil analysis, Thermography, electrical current… to decide about when to perform maintenance task for each key equipment.
General illustration of Equipment conditions versus risks and reparation costs over time are explained in the classical P-F figure (P being the point where the potential failure can be detected and F as the failure point and the goal is to maximize the lead-time between first detection and the failure). The implementation cost of each type of maintenance is indicated for illustration only.
Obviously, Predictive maintenance implementation costs are higher due to additional sensors to be implemented, periodical vibration spectrum analysis or detailed oil analysis…, generally performed by specialized companies, but the overall maintenance cost is reduced, mainly by anticipating major failures (avoiding higher cost of repair and operations loss) and avoiding systematic equipment repairs/replacement (as this is the case for Preventive maintenance).
II- A.I (or machine Learning) solution for maintenance Optimization
During, last decades, under the new era of industry 4.0 and the huge improvement of data science and reduced costs of new technologies, Predictive maintenance is now evolving towards the use of advanced software based on Machine Learning algorithms, which could learn from equipment historical data taken from normal operation period. Once such models are established, companies can predict abnormal equipment conditions, at a stage where even best operators could not detect as parameters seems almost normal. Only a complex analysis of all equipment data using algorithms, could detect a new abnormal pattern which is then analyzed by company’s Subject Matter Experts (SME) to escalate to the effective root cause and decide about anticipative maintenance, either at the soonest time or the at next close scheduled maintenance or plant overall.
If classical Predictive maintenance allows the anticipated identification of failures at its early stage, the use of Machine Learning techniques could raise a warning a step further at even an earlier stage, hence improving the predictability of a major failure and mainly offering a way for an informed maintenance cycle optimization.
Case study example, Predict-it software:
Many success stories are reported following the use of advanced prediction software such as Predict-it (from ECG) which allowed AGL (Autralian major Energy operator) to optimize 19 Million AUD in 3 years and >50 Million AUD through major failure avoidance. Please refer to the link:
Predict-it is an easy-to-use software, developed by ECG (in Ohio) and is widely implemented in Power generation in the USA, Japan and Australia.
Other type of models are those based on Physical first laws and which can be combined with A.I models to identify performance degradation in a complex process. A dedicated article will be shared later for deeper discussion about this topic.
III- Cost of Advanced Analytics Tools
As shown in the graphic, there is a cost of implementing advanced analytics as well as all types of sensors and dedicated highly experienced operators. but it is worth emphasizing that modern factories are normally all equipped with a complete set of sensors for a close monitoring by operators, hence, the additional costs are limited to the data acquisition platform and the purchase/development of Advanced Analytics Software in mainly a “one shot investment”. Operations costs are for the most part related to Human resources and licenses renewal/maintenance. The overall investment cost can be reduced through adopting some existing cost effective solutions as an alternative to expensive standard ones. We will cover in a dedicated article on how to avoid some pitfalls and optimize the overall cost of these advanced analysis platforms, both for both large and medium size companies.
Reinforcing Predictive maintenance by the use of advanced software will be of a real value to classical predictive maintenance tools. In reality, this is the combination of Machine-Learning Models alerts (about abnormal parameter), with the outcome of usual predictive maintenance reports (vibration spectrum, oil analysis, eddy current, electrical current…) with a deep analysis conjointly by plant SME, key operators and monitoring team which could bring the maximum value and converge towards a real payback of such advanced platform. Said differently, companies should be fully committed at all levels to maximize the synergy and expect a real return of investment.
As stipulated by many CEO’s, enrolling in Industry 4.0 is not a choice anymore, but a necessity, and an early enrollment will allow a familiarity with such advanced tools and company-wide learning and an adaptation to this new era. Using IIOT sensors is yet a further step with a standalone components embedding smart code, and is becoming the next move for future modern facilities.