Predictive maintenance is high on the list of priorities for manufacturers, but a recent survey suggests there is still some resistance to a fully-fledged digital revolution, as Michelle Knott explains…
The rapidly accelerating rollout of the Industrial Internet of Things (IIoT) and Industry 4.0 should be paving the way for an explosion of predictive maintenance in the process industries.
Recent research has tried to gauge how well the reality is living up to this ambitious vision of a world where asset management schemes based on just-in-time predictive maintenance can slash costs and downtime simultaneously.
It’s certainly true that data on the condition of assets is increasingly available from right across process plants. What’s more, machine learning is spawning smart solutions that can turn that data into useful information automatically, rather than having to deploy armies of data specialists to analyse the data manually. In spite of this, the new survey suggests that progress has been modest so far, leaving many operations and maintenance (O&M) personnel yet to be convinced.
The study was carried out by researchers at Emory University (based in Atlanta, Georgia) in cooperation with machine learning company Presenso. The Future of IIoT Predictive Maintenance study was designed to identify gaps between the high-level strategic and business drivers of change and the reality of implementation.
O&M professionals do recognise the long-term value of Maintenance 4.0 but are more sensitive to the practical implementation hurdles
Eitan Vesely, co-founder and CEO, Presenso
With this in mind, the team interviewed maintenance and reliability professionals responsible for predictive maintenance within their organisations. They found that there’s a growing chasm between the potential for predictive maintenance in line with Industry 4.0 (PdM4.0) and the reality in many industrial plants.
“We found no urgency to upgrade legacy maintenance and reliability practices from the 1970s and 1980s. Microsoft Excel is still the default analytics tool,” writes lead author Arnav Jalan. “Concerns that are raised about PdM4.0 and Maintenance 4.0 stem from practical considerations regarding the feasibility of deployment and the lack of resources. O&M professionals view PdM4.0 positively but expect an incremental change in the form of improvements to existing systems and processes.”
Incremental change is a far cry from the promised 4.0 revolution. The researchers started by asking O&M professionals what they’re currently using and how happy they are with their maintenance management tools. “Despite the promise of PdM4.0, there is little discontent with current predictive maintenance systems. Traditional predictive maintenance, including vibration monitoring, oil residue analysis and thermal imaging, still dominates, and manual statistical modelling such as Excel has not been replaced by more advanced technologies,” reports Jalan.
Eitan Vesely, co-founder and CEO of Presenso, believes there are several reasons for this apparent complacency among those working at the coal face, even though senior managers are enthusiastic: “Most O&M professionals are already invested personally in their current predictive maintenance systems so there is a natural reluctance to change, especially to those on the frontline. In reality, O&M professionals do recognise the long-term value of Maintenance 4.0 but are more sensitive to the practical implementation hurdles.”
In practise, this leaves many of those surveyed relying on manual statistical modelling tools such as Excel (44% of respondents), which contrasts with 23% who reported using advanced statistical modelling and 12% who use machine learning.
“Why is Excel so prevalent? The reason is basic,” explains Jalan. “Reliability and maintenance professionals already use Microsoft Excel as opposed to other tools, such as MATLAB. Familiarity with Excel creates momentum; people start with a small table that grows exponentially and becomes the default due to lack of alternatives.”
“Another issue is that even if someone is an engineer, their experience with Big Data is almost non-existent,” adds Vesely. “A university-level statistics course is not going to help navigate through the complexities of machine learning.”
The upshot is that there is a general acceptance of the status quo, with most survey respondents saying they are relatively satisfied with the current options for predictive maintenance systems. Although only 2% were Very Satisfied with current systems, almost half (46%) were Somewhat Satisfied and an additional 28% were Neutral. A minority of only 8% were Very Dissatisfied.
We found no urgency to upgrade legacy maintenance and reliability practices from the 1970s and 1980s. Microsoft Excel is still the default analytics tool
Arnav Jalan, lead author, Future of IIoT Predictive Maintenance
It’s tempting to lump the solutions and advantages of Industry 4.0 together and imagine them being rolled out in parallel. However, this research suggests that would be an over-simplification.
“There is no one common definition for Industry 4.0” says Vesely. “I think industrial plants are going after the low hanging fruit. Many so-called Industry 4.0 processes are simply best practices. These are easiest to implement, which is why O&M employees tend to gravitate to them. These include automated scheduling and reporting.”
Research found the incremental adoption of the automation of processes and workflows is the most anticipated change in the near term. Not surprising given that Computerised Maintenance Management Software (CMMS) is familiar territory for O&M operatives. So 45% of respondents expect full deployment of automated failure reporting.
Enthusiasm drops as proposed solutions become more ‘exotic’. Over a third of respondents do not expect to adopt roboticsassisted repair and the prospects for the deployment of the digital twin approach to asset management remain even more remote.
A digital twin is a replica of physical assets. This virtual model can be put to a variety of uses, including learning and updating itself in near real-time to give plant operators a detailed picture of the current status, working condition or position of assets. Some 61% of O&M professionals claim that they are not even familiar with the concept, while a further 20% of respondents do not expect a digital twin to be deployed within five years. Only 4% of respondents expect complete deployment.
A preference for more familiar solutions is unsurprising; most process operators look for the safest bet in terms of risks and rewards. Still, predictive maintenance is top of most shopping lists, says Vesely: “Many executives view predictive maintenance as the highest priority [for Industry 4.0]. Reducing unscheduled downtime by even a couple of points can directly impact the bottom line… and that is what is motivating executives.”
Automated machine learning overcomes obstacles
Israel-based Presenso announced the production release of its machine learning-based solution for predictive maintenance in July.
The solution automates machine learning processes and provides a Software as a Service (SaaS) approach that requires no interaction with the plant’s engineers and no data scientists to perform the application engineering tasks. In other words, it effectively overcomes many of the obstacles that have prevented traditional predictive maintenance and asset management solutions from being adopted more widely.
Presenso has spent two years researching and developing its Auto-ML (Automated Machine and Deep Learning) solution, which has 11 patents pending. A beta product was launched in early 2017 and deployed at multiple customer sites. Presenso says it already has customers across a range of industries.
The new system collects immense amounts of data at very high speed from hundreds of machines and thousands of sensors and streams the data to the Cloud in real-time. Using proprietary deep neural-network architectures, Presenso’s analytic engine autonomously interlinks events with components within the machines and ultimately predicts evolving failures. In addition, it provides information about the remaining time to failure and its origin within the machine.
