If it ain’t broke, the saying goes, don’t fix it. While that approach might work just fine for some situations, today’s modern manufacturing operation isn’t one of them.
In the connected factories becoming more common in the age of Industry 4.0, one downed machine can wreak havoc on an entire production line. According to Deloitte, ineffective maintenance strategies can reduce a plant’s productivity by as much as 20%. Even in older, traditional factories, the failure of a single machine or part can result in collateral damage to other areas of the wider system, such as a failed capacitor frying a circuit board, which then needs to be replaced.
A predictive maintenance (PdM) strategy bolstered by artificial intelligence (AI) applications and machine learning (ML) leverages the volumes of data generated every minute of every shift to help the maintenance team forecast the estimated time to failure based on performance history. Employing such a strategy has considerable upside for manufacturers. According to Deloitte, predictive maintenance programs can increase equipment availability and uptime by as much as 20%, shave overall maintenance costs by 5% to 10%, and result in an eventual reduction in maintenance planning by up to 50%.
Predictive maintenance starts with four questions:
Historical data, which is constantly gathered by the sensors and software running modern manufacturing equipment, can provide answers to each of these four questions. The problem, however, lies in the sheer volume of that data. With hundreds or potentially thousands of points to monitor and track, it is not humanly possible to manually process it.
Enter AI and ML. These tools not only tackle the reams of data collected, but also can ensure their accuracy. AI models can generate these predictions in near real-time, which is key to avoiding undue downtime or injuries to human workers.
The first step is to compile a list of all the equipment in a facility, then determine the historical costs of maintenance, repairs and downtime as well as assess the risk if a machine is not maintained properly. Once this cost/benefit ratio is calculated, select a proof of concept that provides a decent ROI without an extended ramp-up period to get started.
There are two types of data that can be leveraged for predictive maintenance: required data, without which the AI algorithm will not work, and supplemental data, which can provide additional context and improve the overall quality of the resulting insights.
The data science team then rolls this data through AI algorithms to build a predictive model to determine the exact data points — the specific sensor, reading and/or threshold — that is the common indicator of a pending problem. Once the system knows what to look for, impending issues can be flagged as soon as they are detected, avoiding bigger, more costly repairs that would have been required had the issue not been detected.
In addition to the financial and productivity benefits, there are other, less obvious ways that a PdM strategy can help a business's bottom line.
The benefits of implementing an AI-driven PdM strategy are many, from reduced equipment downtime, to slashed repair and maintenance costs to increased worker safety, and the investment in time and resources pales in comparison.
Asghar Ghorbani is lead solutions engineer at H2O.ai.
Timely, incisive articles delivered directly to your inbox.