Worldwide competition is putting today’s corporations under intense financial pressure and operations and maintenance budgets are among the first to be slashed.
Maintenance can be classified into three categories:
- Reactive – after the machinery has failed
- Preventive/Scheduled – at a fixed schedule irrespective of the current health of the asset
- Predictive – to be able to predict the time to failure and accordingly carry out the maintenance activity
Market research shows that 86pc of maintenance is reactive (too late) or scheduled (unnecessary) and typical maintenance practices have not changed much over the years. This could be attributed to the unavailability of tools powerful enough to fundamentally improve maintenance practices.
At present, the most asset intensive industries follow scheduled (time/usage based) preventive maintenance practices. This practice still does not eliminate the possibility of unscheduled maintenance and catastrophic failures. So companies are trying to make their operations reliable and ensure optimal performance at a lower maintenance cost by predicting and preventing failures in a timely manner.
As per The International Society of Automation:
- 5pc of plant production is lost annually due to unplanned downtime
- $647 billion is lost globally by manufacturers across all industry segments
With IoTizing of assets and advancements in Machine Learning and AI, it is now possible to carry out predictive maintenance of assets. This can significantly bring down the cost of assets. It is applicable to both assets and heavy equipment out in the field as well as critical machinery on the shop floor. The only difference will be in the methods to collect data, but once the data is obtained techniques to arrive at predictive maintenance remains largely same.
Typically predictive maintenance is carried out by identifying the key performance measures of health/wear and tear of an asset. These could be, for example, vibration, temperature, load, inclination, etc. These performance measures are then constantly monitored using sensors and the data collected is monitored over a period of time for anomalies to predict the health condition of the asset.
The cost savings/benefits resulting out of predictive maintenance manifest themselves in many ways:
- Cost savings in maintenance of the asset – Being able to predict the health of an asset on a continuous basis and avoiding the OpEx spent on regular scheduled maintenance can bring down the OpEx by as much as 60-70pc. E.g. being able to predict the health of a rolling stock can avoid bringing the rolling stock back to yard for regular scheduled services, this can result in significant cost savings to the rail services as it can now run those many more trips.
- Downtime or failure avoidance of the plant or machinery – Many of the machines in discrete or process manufacturing run critical systems and even an hour’s down time can result in losses of millions of dollars to the corporation. e.g. Down time of agas turbine within a thermal power plant can impact the power production in the grid and can result in losses not just to the thermal power plant but other essential services that run on electricity. Hence, it is very important to be able to predict the health of these machines within the factory floor to avoid such losses.
- Inventory and Supply chain optimization – Often, the plants are tasked with lean inventory to optimize the cost of maintaining inventory. However, when a particular machine part fails the replacement spare part of that particular make and model may or may not be available in the stock resulting in elongated down times. Being able to predict the failure of the machine and the root cause down to a particular machine part can greatly benefit in terms of ensuring that the spare part is available in inventory in time for the replacement. Varying degrees of automation exist for such scenarios, from the very basic of just alerting the inventory/procurement manager of the impending need of a spare part, sending alerts to near-by locations in case they have a replacement available, integrating with supplier enterprise systems to raise a purchase order for the spare part.
- Newer SLA based business models – With GE now calling itself a digital company armed with initiatives such as Brilliant Factory, the world is now moving towards ‘product as a service’. So rather than selling the equipment/asset, corporations are now selling the fixed performance-outcomes from the asset. They are then charging the customer for the usage of the asset, converting the CapEx into OpEx. The customers have to no longer spend millions of dollars upfront for procuring these machines, rather they can just pay for the usage and the equipment manufacturer will ensure that the machine delivers the necessary agreed performance SLA. For such scenarios it is in best interest of the equipment manufacturer to be able to constantly monitor the performance of the asset and predict any type of downtime or anomaly that might impact the performance of the machine.
The Indian Automation and Manufacturing Scenario
Whether you look at England, the United States and Germany, or, more recently, Japan, Taiwan or South Korea, no country has climbed the economic ladder to prosperity without many years of manufacturing-driven growth. To address the manufacturing challenge, the previous government in India set up a National Manufacturing Competitiveness Council which started new programs focused on quality and clusters. The incumbent government has taken the focus on manufacturing one step further by starting a national Make in India campaign. This has come at a time when a lot of product companies arising out of India are competing at a global scale and technical know-how is available courtesy India’s software technology pool. Hence, coupled with state-of-the-art automation systems, companies are now leveraging the talent pool to build predictive maintenance systems leveraging technologies such as Cloud, Big Data and Machine Learning.
What is the solution?
The solution typically involves edge sensors which would send data to a gateway device with IP capabilities. The sensors keep sampling the performance measure connect with the gateway over some low power connectivity mechanism such as LPWAN, BLE, ZigBee etc. The gateway should ideally possess some edge computing capabilities to aggregates this data and selectively send it to the cloud. Edge analytical capabilities in the gateway are preferred to avoid having to send every small data to the cloud and take some decisions on the edge itself. It should be able to detect some basic anomalies and alert the field service operator. The data on the cloud is used to throw up dashboards around trends, anomaly alerts, overall health of the system etc. This data is then stored in BigData systems and used for further analysis using Machine Learning algorithms and techniques such as regression, clustering etc.
The solution for a predictive maintenance use-case typically involves multi-disciplinary teams with exposure to various aspects such as mechanical engineering, instrumentation, connectivity protocols, cloud and machine learning.
Data from past trends can be leveraged for machine learning where analysis of trends can be leveraged to predict the future trends and thus being able to predict anomalies that a particular asset might develop over a period of time. There are also some frameworks available like that of Azure ML which provides out-of-the-box template for providing input data and getting RUL (remaining useful life) and TTF (time to failure) for the asset.
While the path has been laid, it remains to be seen, whether the organizations, which implement these techniques most effectively the fastest, will be able to reap benefits and come out as clear differentiated winners.
For More Information Visit: http://www.sasken.com