Predictive Maintenance - not a theory but a practical application to produce tangible benefits
Among the most frequent causes of production stoppages, and thus of a company's loss of tangible and intangible resources, plant failures are a critical issue that can occur in any industrial setting.
Although the certainty of being able to avoid them altogether is remote, the introduction of predictive maintenance, compared to the reactive maintenance employed until the 1980s, has numerous advantages referring to the optimization of machines and processes.
It is, in fact, a Smart Maintenance procedure that makes it possible to predict the risk of anomalies and plan the interventions necessary to curb or avoid failures, as well as to encourage the training of skilled labor and streamline parts replacement operations.
In addition, the combined use with Industry 4.0 technologies increases the benefits from this practice, as they further simplify the issuance, collection and understanding of meaningful data for efficient machine management.
An example is the case of a customer operating in the plastics industry, whose production performance was affected precisely because of a crucial system-wide plant failure that occurred about a year and a half ago. Specifically, the damage affected three water filling pumps, resulting in a total shutdown of the medium. The impact on production was significant, as onerous intervention costs were compounded by economic losses due to the interruption of the work cycle. In this case, in fact, the prolonged MTTR, i.e., the average repair time, hampered work shifts forcing the stoppage of labor.
Since this was an issue encountered other times in the past, the customer had already taken a reactive approach, trying to improve maintenance by switching from a fix when fail policy-whose purpose is only corrective-to a preventive one, based on planning the interventions to be addressed to avoid the occurrence of failures. However, despite the decrease in the number of damages suffered by the plant, the limitations found by this change were mainly on the economic side of the business. In other words, planned stops to act on machinery and prevent breakdowns continued to affect costs, making them excessive and difficult to sustain.
Hence the management's decision to turn to the Automate team for the purpose of defining and developing in synergy a predictive maintenance project thanks to which it would significantly improve the management of machinery with a strong focus on cost containment and at the same time ensuring full production uptime and above all reduction of risks such as to produce negative consequences for the business.
The first step in this direction was the company's initiation of a fully autonomous offline data collection prior to the start of the project. This was a smart choice as it was functional in improving the readability of the machinery and categorizing problems encountered over time. In fact, the same data provided a useful base from which to build a quality solution, the value of which derives mainly from the ability to handle complexities of various kinds. On the one hand certainly that of collaboration between different skills, but without forgetting the need - which emerged already during the analysis phase - to consider in parallel the integration of a platform for real-time data collection with existing systems and the implementation of forecasting software and algorithms in order to make them flexible, secure and reliable. In order to achieve this, in addition to the work of selecting the most suitable Data Analysis and Machine Learning methodologies, it was deemed appropriate to involve management in all the choices related to the different project phases, so as to reinforce in parallel the client's awareness and autonomy.
Acting in this way also made it possible to better address the issues that arose during the first moments of analysis and design of the technology architecture. For example, having to respond to a discrepancy of information produced by the data collected by the company - fragmented and not entirely consistent with the dynamics of failures - or the difficulty of imagining in advance the real usefulness of the monitoring system to avoid failures and, above all, to plan the necessary maintenance operations.
In the first case, the data collected were read by initiating pre-processing and cleaning actions in order to extract value from them, while the planning of implementations was optimized through constant audits, giving the customer the opportunity to immediately appreciate the results achieved. A well-established approach that is still relevant eighteen months after commissioning of the solution, which, by aligning with the different production needs, has an evolutionary nature for which the intervention of the technology partner has been required at different times: to modify the software, add new sensors, assist the staff in learning some specific indications provided by the algorithm or intervene in other areas of the value chain such as, for example, the presses.
The ability to interpret machine behavior, along with a greater understanding of the causes of failure, has enabled the company to date to reduce the number of repairs needed in a year by 20 percent and the average maintenance time by 30 percent.
Of course, this was a gradual process for both management and employees, so that both could embrace the change with awareness, feeling mastery of the new technologies introduced. Therefore, after an initial analysis and ideation phase, the development of the solution followed three phases, respectively inherent to:
- The integration of the platform for real-time data collection and visualization, aimed at making some complex dynamics accessible and activating an alerting system to control some possible issues.
- The collection of data and the development of the predictive algorithms needed to read them, which in this case made it possible to identify alerts and avoid the recurrence of the malfunction based on the variation of certain quantities from previous experience.
- A further improvement of the Predictive Maintenance method adopted, which proved capable of acting on the three machines by exploiting the information produced in order to predict a type of failure one month in advance. Sufficient time to prevent breakdown and plan maintenance without the risk of hindering production.