The idea that a machine could tell you when it needs maintenance before it fails is quite alluring. Because of that, predictive maintenance has joined the list of buzz words surrounding the Industrial Internet of Things (IIoT).
Although the machine learning and artificial intelligence (AI) solutions that enable predictive maintenance are helpful, they’re only as helpful as they’re trained to be, according to a recent Forbes article titled “Why predictive maintenance is not a silver bullet solution for manufacturers” by Willem Sunblad, CEO and co-founder of IIoT company Oden Technologies.
“While it’s a very attractive proposition, it’s crucial that manufacturers understand the dangers of getting it wrong,” Sunblad says. “A false alarm triggered by inaccurate data can lead to incorrect actions and additional costs.”
In the article, Sunblad details the challenges that inaccurate data can pose to the implementation of an accurate and effective predictive maintenance model. But first and foremost, implementation starts with a solid understanding of process variables and machines, and a strong data set.
“Understanding that some data sets are harder to collect than others will be a huge asset in the decision-making process,” he says. “For example, if a machine only breaks once a year and a hundred observations of the break are needed in order to build an effective predictive model, it is clear that observation will not be feasible.”
Although this is a hurdle, Sunblad notes that some organizations partner with machine-makers who already have a data set, or they’ll find a partner who can create a digital twin of their machine that can collect the data sets.
But notably, the predictive maintenance model will only be able to predict those problems for which it already has data. Unexpected problems can still remain unexpected, Sunblad emphasizes.
“Unexpected problems can still occur, because if a problem has not been anticipated, it’s unlikely that any relevant data sets have been discussed and collected,” he says. “Manufacturers should keep this in mind, because the most expensive failures are often the unexpected ones.”
While predictive maintenance can be helpful, Sunblad suggests that predictive operations and predictive quality are easier entry points with greater amounts of data available.
