Home / Topics / Smart ways of working /

Datadriven maintenance

Data and datadriven methods

Data from various processes and systems generates at an ever-increasing rate and volume. This data paves the wayfor datadriven work and decision-making processes in the production industry. The use of new methods and technologies like data analysis, data visualization and Machine Learning (ML)/Artificial Intelligence (AI) are made possible by data collected via, for example, sensors, control systems, quality and service systems in industrial production systems.

Predictive maintenance

One of many scopes for data driven methods is maintenance where AI can be applied for predicting wear and deterioration of machines and other production equipment. This is called predictive maintenance. For predictive maintenance to be developed, data and measurements from machines collected over time is needed. This can for example be measurements of vibration in a machine with rotating parts or temperature measurements from a production unit that heats/cools products. When this data is collected over time they can be used to train an AI to recognize patterns that corresponds to errors, deviations, or deterioration in the machine/process.

Maintenance principles

A maintenance policy/principle is a rule that decides when a maintenance measure is triggered. Note that maintenance strategy is a broader concept that includes the whole decision chain, not only limited to maintenance measures.

Different maintenance principles are suitable for different assets. Common principles are:

Corrective maintenance
Preventive maintenance
Condition-based maintenance
Predictive maintenance

Corrective maintenance is a policy that means that an asset is left in operation until it has lost its function due to wear and tear and maturity. The asset is then replaced or repaired. This policy applies to assets where the cost of maintaining the asset is greater than the cost of repairing or replacing it. An example of such an asset is light bulbs. The possibility of maintaining light bulbs is limited and replacement is a low cost. Therefore, light bulbs are kept lit until they break. Corrective maintenance has the advantage that you get the maximum service life of the asset before maintenance needs to be done. A disadvantage is that the maintenance work can generally not be planned. Most industrial machinery and other assets are not suitable for this policy due to high demands on stable and reliable operation.

Preventive maintenance is a policy that is based on planned maintenance measures with a certain frequency. The frequency can be based on calendar time or time in operation. This policy is suitable for assets where the cost of maintaining is (much) lower than the cost of repairing or replacing the asset. This is true for most industrial machines and thus a common maintenance principle. The disadvantage of preventive maintenance is that the entire service life of the asset is not utilized before a maintenance measure is performed. A certain number of "unnecessary" maintenance stops occur when the asset does not need maintenance. The advantage is that maintenance can be planned well in advance.

Condition-based maintenance means that production assets are monitored with various measurements to continuously assess maintenance needs. An interval is often determined where measurements that end up within the interval is said to be “normal” operation and measurements that end up outside is “abnormal” operation. When one or more measurements outside the limit values have been observed, maintenance is planned. The hope is that the asset has not lost function but the maintenance measure required is less than that needed if it has completely lost function. Condition-based maintenance as a policy is suitable for assets where the cost of maintenance is less than the cost of replacing/repairing the asset. The policy can also be combined with other maintenance principles such as preventive maintenance.

Predictive maintenance is similar to condition-based maintenance, but instead of rigid intervals that define normal versus abnormal operation, more flexible rules are used based on learned patterns in data. Predictive maintenance as a policy suits assets where data availability is high, and the cost of planned and unplanned downtime and maintenance is high. Predictive maintenance can be combined with other maintenance principles to increase the flexibility of the maintenance policy.

The way towards data-driven maintenance

Data-driven methods is a collection of tools, and like all other tools, they are suitable for different purposes. Data-driven methods presuppose high data availability and digitalisation maturity. The introduction of these methods in their organization is an interplay between technical, organizational and economic factors.

Data-driven methods require the collection, handling, quality control and analysis of data, which in turn may require specialized software/hardware. Data-driven maintenance can also affect the organization by requiring new skills, changing working methods and influencing management processes. Changes in the company's working methods and investment in software/hardware, as well as the development of competence, costs money and it is important to try to estimate if/when the investment will yield a return.

One approach is to follow the motto "think big, start small, learn fast". Start collecting data in a machine that can quickly provide insight into whether data-driven maintenance suits the organization. Maybe the asset already collects measurements via control systems or sensors, and these can be extracted? Iterate and continuously evaluate which type of work method suits your organization.