Reaping the rewards of a sound predictive maintenance strategy

Predictive maintenance is a powerful preventative medicine which can cure the production maladies that impair many industries. To help organisations enjoy the full potency of this approach, Maurizio Canton, VP Customer Success at TIBCO Software, has offered the following advice. 

Predicting and subsequently enacting preventive maintenance for industrial equipment involves anticipating failures and planning controlled responses to minimise any disruption. The sooner faults are nipped in the bud, the better for productivity. The nature of the response will vary according to the problem, which itself is calculated in advance and according to predefined parameters. 

While over-maintenance of equipment is expensive and inefficient, these costs can be minimised and even avoided if the nature of the problem is quantified in terms of the variables that define it. This might be, for example, necessary elements like the oil needed for machinery joints. The volume of oil present and its temperature are two data points that help paint a picture of how an engine is functioning. In conjunction with other relevant data points – for example, stress in the metallic structure – analysts can programme an alarm to activate in advance of a component wearing out or needing replacing. 

This discipline of predictive maintenance represents a massive saving to a company, such as one operating a fleet of helicopters in the North Sea, where all machinery is subject to an environment high in salt water, which is corrosive. You could not expect the same life component cycle if identical machinery were in use in a desert operation.

By recognising the signs typical of machinery about to malfunction and automatically responding, manufacturers can give a quick and effective response that can be incredibly versatile. By catching problems before they escalate and become critical, predictive maintenance pays for itself many times over in savings on time and money. This makes the business case very powerful, cutting downtime by over 50%. The only limit on predictive maintenance’s versatility is the imagination of the programmer. The greater their powers of anticipation, the more downtime they can eliminate and the closer the company gets to full productivity and maximum profitability.

Some strategies employ mechanisms such as machine learning, which is a major step forward, as it empowers the machine with some cognitive capacity. The equipment can learn from previous events and modify its responses. This ‘intelligence’ is then incorporated into the operating system of the device.

This operation involves key steps toward the analysis of the machine’s event history. It does this in order to deduce the rules that will, in future,  be applied in ‘real-time’ in order to trigger, if necessary, the actions needed to maintain the machine and keep it operating. 

The application of predictive maintenance

The best way to execute predictive maintenance is to consider its relevance to the company and identify the benefits. The first step is an audit of the production line: measure the live elements of the equipment – such as accelerometers, pressure sensors and vibration sensors – using the Internet of Things (IoT). Map all available data and study how and where all the vital intelligence about the machine’s performance can be collected.

The second stage is the analysis. This means checking that the data is in line with the objectives. At this stage of the predictive maintenance strategy, companies need to understand the behaviours they are identifying, so they should adopt a data science approach to intelligent analysis. The objective is to unearth significant anomalies. These are highlighted by the use of data visualisation, descriptive statistics and machine learning.

At stage three, companies must use this data to make their failure prediction models, analyse potential outcomes with business experts and experiment with these models. The more consistent the event history, the more reliable the models will be. After validation, these models will be the basis of strategies put into production in the analysis chain, supported by continuous data flows with real-time feedback of sensor information. This completes the administration of predictive maintenance and should bring the system closer to the ideal of ‘perfect’ productivity.

The scale-up and data management challenge

Predictive maintenance is not completely snag-free. Indeed, customers must face the twin challenges of effective data management and scale-up. Companies often make the mistake of skipping some stages in order to focus on the end result. This means that they don’t give enough thought to whether their model works on a bigger scale. The principles that apply to a single machine may not be accurately reproduced across an entire fleet as complexity can multiply with each extra unit added. 

In this case, it becomes necessary to use a scalable system that can be upgraded on a wider scale, without jeopardising its accuracy or potency. Predictive maintenance often works very well at ‘lab scale’ or in a specific project, but it cannot be reproduced en masse across a fleet of 100 machines or a group of plants because those aggregations can become too complicated. 

In addition, companies must learn to manage their data well. They should never embark on a project without having assessed their data capital. They need to identify the information they require, as well as the gaps in information and they must understand the relationships between these types of data. Finally, a very detailed knowledge of your business issues is the key to success in predictive maintenance.

Multiple factors need to be considered when employing predictive maintenance. Companies can justifiably expect success, but they have no hope without careful monitoring of the decisive steps, the close involvement of business teams and a detailed roll out plan across the entire estate.