The value and benefits of predictive maintenance
What is the value of digital preventive maintenance solutions? Or how expensive is downtime due to device failure nowadays? Machine manufacturers and also operators increasingly want more insight into the machine status and especially into the predicted machine status.
In recent years, the technology of intelligently networked machines has greatly simplified, and specific use cases are already proving in numerous industrial sectors that operating costs can be reduced through predictive maintenance and that product and process quality become more sustainable. Digitization therefore serves exactly the requirements in order to be able to continue to play a leading role in global competition in the future.
Benefits of predictive maintenance
Customers are looking for smart, digital products that support their preventive maintenance program with the following benefits:
Reduce parts costs and capital expenditures. By reducing service costs and increasing the overall life of a machine, long-term capital expenditures are reduced. In addition, due to the increased confidence in the operating condition, a reduced stock of spare parts is possible.
Increase operational efficiency. Keeping machines running longer between service intervals and having a high probability of future problems can improve overall operational efficiency. This leads to better business results and a competitive advantage.
Reduce ongoing compliance costs. Corrective and preventive maintenance is costly from a labor law perspective. Improving the visibility of problems before they arise can have a significant impact on lowering the ongoing compliance costs associated with them.
Increase the product and process quality. Since machines work more efficiently and with a higher quality and service life, the resulting quality and effectiveness of the processes, which depend on the machine performance, also increase.
These are just a few of the benefits of predictive analytics as part of a preventive maintenance program.
By adding more sensors to devices and in combination with an IoT platform that can collect, process and visualize large amounts of machine data from multiple machine types, locations and applications, the following additional benefits can be achieved:
Real-time monitoring of machines. Local analyzes at sensor or device level, with down-sampling of high-frequency sensor values and aggregation in the edge to reduce latency times, bandwidth and storage costs in the cloud, with simultaneous transparency of machine performance.
Stream analytics. Analyzing data, transforming it and examining it for patterns and trends that indicate certain problem areas can diagnose faults in the machine very quickly. This includes the display of the machine status, instructions for predictive maintenance and service provision.
Flexible options for data visualization. An IoT platform can allow a visualization adapted to the respective user (and his role). This is of particular interest to manufacturers of networked industrial systems who want to offer their end customers a specific brand experience.
Analysis for multiple machines, regions and applications. By collecting data from multiple types of machines in multiple locations and across multiple applications, a wide variety of data can be collected and analyzed using big data tools to identify trends that may not be apparent by looking at individual machines.
Data analysis: a maturity model
In recent years, digital data models have increasingly come onto the market, especially for rotating machines such as pumps and motors. Nevertheless, the industry in the broader sense of data analysis is still at the beginning. As a result, companies still have a significant opportunity to differentiate themselves from the competition by developing and deploying pragmatic data analysis solutions.
In addition, it has been shown that especially those companies are particularly successful that proceed in small development steps and thus continuously build their knowledge progress. In all cases, this process begins with the basic connectivity and the combination of employees' existing knowledge of machines or components. The next step is to combine the knowledge with insights into what exactly needs to be measured. Now you can try to answer questions such as: "How do we automatically determine when an elevator will fail within 90 days with a probability of 60%?"
Only after these intuition-driven hypotheses have been developed can data analysis be applied to real machine data in order to prove or disprove it over a certain period. The combination of human input and intelligent automation is one way of predicting a machine failure. As an alternative, so-called "unsupervised machine learning" algorithms are available today, which independently derive appropriate data from various data. In most cases, it is possible to switch to operational operation after a short, automatic training phase for the algorithm. The probabilities of a correct prediction in such systems are> 90%.
Best practices from leading manufacturers
Here are a few more best practices from some of the leading users of predictive analytics technologies:
Rely on the human intuition of experienced mechanics. Do not discard this important human knowledge, but rely on it. Leading companies entrust their top mechanics or machine designers with predictive maintenance projects who understand the operation of the machine and know under what circumstances certain components often fail.
Crawl, run, run (crawl-walk-run). Don't succumb to the false truth that finding insights for predictive analysis can be completely automated. While trusting experienced mechanics to hypothesize is a critical first step, the nuances that can lead to a machine failing can surprise you. Start slowly, collect data, prove / disprove hypotheses over time and then find the right mix of sensors and report frequencies as a learning process that evolves over time.
Use predictive analytics as a trip. To begin with, they find themselves in a causality dilemma when it comes to determining which comes first: the data or the analysis. In reality, the two are inextricably linked. Understanding predictive analytics as a journey that begins with basic connectivity is a key recipe for success.
Conclusion
Predictive maintenance programs don't have to be costly. Especially when companies rely on human intuition to develop hypotheses, i.e. follow a crawl-walk-run approach and view predictive analytics as a journey. You start with the basic connectivity and get a steadily more precise picture through continuous data aggregation and analysis. On the Lean.IQ marketplace, we present some specific concepts that can significantly accelerate implementation. Leaders in it, applying these best practices can turn costly operating costs into a source of significant competitive advantage.