Why Predictive Maintenance with AI, ML and IIOT is the trend for future

The process of maintenance over the years has evolved based on the circumstances and the technological advancement in the manufacturing process. When the manufacturing process started to invest in machineries and equipment to take over from the manual process, the maintenance concept was limited to maintain the equipment once it breaks down. Over the years as manufacturing process started to use new technologies and increase in production activity, the maintenance concept shifted to planned preventive maintenance. The objective is to achieve maximum efficiency (OEE) in production and to drive major cost savings in maintenance.

Today, preventive and predictive maintenance planning is the norms in industrial equipment maintenance process and it is a standard condition to stay competitive in the market. 

To execute any maintenance strategy several software tools help to reduce the manual errors, lethargy in scheduling of maintenance activity, creating maintenance notifications and recording automatically all maintenance information with the aim to get the maximum Availability of the machinery. 

Planned Preventive and Predictive Maintenance
Preventive maintenance and predictive maintenance are the two major maintenance process adopted across industries currently.

Planned preventive maintenance follows a schedule driven by time (calendar and machine time), events or production schedule. The equipment is in operation and necessary maintenance is planned to reduce the downtime, cost of maintenance, resources and increase the productivity. Necessary resources, spares and consumable materials have to be planned in advance to perform the maintenance activities. On the scheduled maintenance day, the team carries out the maintenance, records the work done, spares replaced and gives clearance for operation. 

This method, though effective, has a drawback of over maintenance or under maintenance based on the current condition of the equipment and does not assure that the equipment will be available for continuous operation free of any breakdown until the next scheduled maintenance. Probability of increase in the cost of maintenance, if the planned maintenance was not required on the day it was performed or the breakdown occurred ahead of the scheduled maintenance plan cannot be ruled out.

Predictive maintenance on the other hand uses condition monitoring, vibrational, thermal, visual and acoustic feedbacks from the equipment for the users to plan the maintenance requirement. Based on these monitoring activities an alert is generated in advance to take proper corrective action to avoid any major breakdown. IIoT (Industrial Internet of Things) sensors are used in the systems to provide real time data, which are connected to software for maintenance (for example a CMMS) to analyse and schedule the predictive tasks on the machinery. 

The accuracy is limited as Scada and CMMS system are not capable of analysing the real time complex dynamic data from the equipment and extrapolating these data for predictive maintenance as these systems are manually programmed in advance to take actions based on limited feedbacks visualised by human intelligence.

An example to understand is given here. Take the case of engine of a vehicle alerting for oil filter clogging. This data is a good indicator for taking it up for maintenance, but then there could be other parameters and conditions to be analysed for the filter clogging. After maintenance repeated alert of filter clogging cannot be ruled out and the repeat cost incurred. 

With the advancements, came the era of Industry 4.0 with Artificial Intelligence (AI) and Machine Learning (ML) process which is a subset of AI. AI and the IIOT (Industrial Internet Of Things) has further revolutionized the way today various solutions are arrived at. The trend now is in the use of AI and IIoT in the field of maintenance to achieve the maximum efficiency (OEE). The future of maintenance of equipment lies in the use of AI and IIOT. 

We can define this new way of maintenance as Prognostic: predictive of something in the future. 

Predictive Maintenance with AI and IoT 

Going ahead AI will chart the trend in Predictive maintenance process along with machine learning. How? Manufacturing Industries, over the years have gathered massive data about equipment, process, breakdowns which are used to further aid in Predictive maintenance process.

Gathering data about the operational procedures, manufacturing process, informational data will be processed and analysed through Machine Learning. Historical data about a particular object/condition from maintenance records, IIOT based sensor data can be analysed for its operation and deviations from the normal. Real-time data processing and automation will make use of historical data to gain better insights into the root cause of machine failures. They recognize failure patterns, and predict when malfunctions may arise based on the similar conditions under which they arose in the past.

The following Data sources are collected for Predictive Maintenance:

1. History of parts/equipment failure – History of data during normal operations and failures are inputs for ML algorithms; 
2. History of Maintenance/Repairs – This provides records of parts changed/repaired, occurrence during the operation of equipment;
3. Records/logs of Machine operating conditions – sensors provide real time data during normal operation, abnormal conditions and failures for the algorithm to collate with other historical data to predict failure, degradation or alarm about any anomalies in the operation.

This will more accurately predict the time or nature of maintenance required. AI can aid in taking a real time action faster than human intelligence to keep the critical equipment in operation. AI analytics and commands serve the specific function of optimizing maintenance cycles for industrial machinery. 

Manufacturing system with AI predictive maintenance can extend equipment life, improve efficiency among field technicians, and significantly decrease expensive machine downtime.

All factors that greatly minimize production delays and hazardous work conditions that could cost manufacturers a huge hit in their revenues. Issue of over maintenance, where costs are wasted on the work time spent changing or repairing parts more frequently than necessary, as well as the unused portions of parts that go unaccounted for in preventive maintenance model will be avoided.

The advantage in using Machine Learning is it can dynamically adjust to new data sourced from various inputs and with the algorithms programmed will process in real-time the corrections to be taken, also detecting and alerting staff of serious issues. This avoids the need for manual configuration, data selection, or threshold settings that other maintenance measures demand.

Referring to the earlier example of vehicle engine oil filter clogging alert, the predictive maintenance with machine learning can analyse other parameters and identify the cause of repetitive filter clogging. It could be quality of oil, engine wear and tear increase, sufficient coolant not available etc..

Why is Predictive Maintenance with AI, ML and IIOT the trend for future?
Because it can predict to not run maintenance too early and incur cost on the work that was not needed, and also not delay the maintenance on the equipment and avoid a breakdown state of the machine with more accurate prediction. 

The future of maintenance will be focussed more on proactive maintenance as prediction of time of failure of machine will be available and action to be taken can be planned to avoid the down time.

Machines equipped with AI predictive maintenance make use of real-time monitoring to understand the exact condition of a part at any given moment. The failure time of machine and its parts will be self-diagnosed and actions to be taken in respect of procurement of spares and technician’s required action will be notified by the system itself with minimal human intervention. The smart system will instruct technician to take action only when required and where it is needed. 

By extending component lifespans (compared to preventive maintenance) and reducing unscheduled maintenance and labour costs (over corrective maintenance), businesses can gain cost savings and competitive advantages. 

The maintenance process with digitization and automation will definitely aid maintenance activities optimisation but at a cost. Are industries ready to invest in this? Probably not everybody. 

Here comes the opportunity for another business venture – Maintenance as a Service or MaaS. Equipment manufacturers have the chance to leverage their maintenance business as all data needed will be available on cloud and does not need a large team take care of a vast number of installations across remote locations. Customers would be more than happy paying a premium price to access the maintenance service by equipment manufacturers, than investing in one at their cost. And it is already being practised by some OEM’s like Thyssenkrupp Elevators. They have real time data access across majority of their installations and the failures are predicted and technicians alerted all by the system itself.

For these reasons Predictive maintenance with AI and IIoT is used in various processes and more so in Manufacturing and Automotive industries as early prediction of failure of parts will have repercussion on their business.



Will Predictive maintenance with AI be effective for all business?

To presume that the future of maintenance using digitization can be predicted for all business cases may be not true. Predictive maintenance demands certain qualifying criteria to be applicable for particular business process. Some of them are listed below:

• The problem or failure related to any equipment/machine which can occur has to be predictive in nature. The problem or failure should have an action which when taken will prevent failures when they are detected;
• The records of the operational history, maintenance, replacement, error reports, operational logs, repairs logs related to the problems of the equipment should available to qualify for Predictive maintenance;
• The requirement for Machine Learning-based approach is Data. The data should be relevant, sufficient, and quality data to build effective predictive models and patterns using historical data. These will then provide higher accuracy in failure predictions for predictive maintenance use.

As the data becomes more and more fundamental, machinery connected with OEM data centers will help to increase the proper use of machinery, unleash the full potential of the equipment and thus improve efficiency, reduce costs and increase competitiveness. But what characteristics must the data have in order to be considered useful to be used to achieve the above noble purposes? At least 3 that I consider important here below:

1. Relevant - The data that is collected to analyse a problem has to be relevant to that particular problem only. Take an example of a motor. To study the failure of bearings, features of data relevant to bearing has to be collected. Whereas the if failure of motor has to be analysed then data of all components constituting a motor has to be collected to arrive at a model for Predictive maintenance.
2. Sufficient - Here more the number of failure events records that is available will make the predictive model more accurate. How much is sufficient data has no definitive numbers, but depending on the events and the experts who work on it decides the sufficient data for the predictive model.
3. Quality -The quality of the data collected decides the accuracy with which targets of failures can be predicted.

A report from Deloitte illustrates how titanium-cutting machinery uses embedded vibration sensors and torque monitors to predict the optimal instance when the machines diamond-tipped blade needs to be sharpened. The level of dullness of the diamond tips and, thus, the optimal time to sharpen them, has been difficult to figure out because of many different variables that affect it. The use of vibration or sound sensors and torque monitors can help assess the state of the machinery, as dull tips move and sound differently. The cost savings can run into the millions, as the cost of an hour of downtime went up $260,000 per hour, on average, between 2014 and 2016.

References
To sum up the above insights on the future of maintenance, several reports are available on the internet on how predictive maintenance, IIOT and AI are revolutionizing the way to perform maintenance of machinery.