Predictive Maintenance - Optimize Holistically with AI

More and more companies are therefore turning to predictive asset management to analyze complex asset systems precisely and in real time and to operate them in an optimized manner. In practice, such approaches are limited to solving individual aspects. However, only holistic system solutions offer added value and thus create the basis for a successful predictive asset management strategy.

Idle Machines Cost Money

When planning maintenance and repair, there are various challenges that need to be balanced on a daily basis. If a machine is at a standstill, it devours money every minute. However, it is also clear that over-maintenance causes unnecessary costs due to strict maintenance cycles. Consequently, it is necessary to balance high availability with minimum maintenance requirements. This challenge becomes greater the more machines are in operation. This is because the number of influencing factors increases with each asset, some of which are mutually dependent or mutually exclusive.

In this balancing act, many companies rely on a forward-looking strategy in which optimized maintenance and servicing decisions are made by continuously monitoring the condition of the machines.

AI-based Methods Prove their Worth in a Holistic Approach

Solutions that not only take into account technical data, e.g., pressure, temperature, or hours worked since the last maintenance, but also include business aspects such as adherence to schedules, utilization of resources, state of depreciation, or need for modernization in the decision-making process-in a cumulative and balanced manner- have proven particularly successful. Due to the volume of data and complex interactions, this is achieved primarily by AI-based methods.

Flexibly Scalable - Whether One Asset or Geographically Distributed Asset Networks

Qualicision's field-proven AI-based, self-learning decision support and optimization continuously evaluates different assets based on qualitatively-labeled asset data - with flexible scalability. This makes it suitable for both predictive maintenance of a single asset and predictive asset management for geographically distributed asset networks.

This creates an additional, AI-independent explanation layer whose simple visualization makes the system's decisions comprehensible and usable even for non-data analysts. The basis is provided by Qualitative Labeling (see Figure 1).

Using a corresponding labeling function, the software observes, for example, which temperature ranges of the sensor data provided indicate a need for maintenance and differentiates between positive, i.e. more desirable, machine states and negative value ranges, i.e. undesirable machine states. It then assigns positive and negative connotations-the so-called labels to the corresponding sensor data.

Easily Comprehensible Visualizations

The software establishes interactions between the determined labels and recognizes patterns in them, from which it derives short-term, medium-term or long-term maintenance recommendations (maintenance labels). Once defined, the labeling functions process and connote any signal sequences.

By visualizing the labeled data in impact and interactions matrices, the software allows users to easily understand the derivation of recommended actions and interactively operate the system.

For the machine monitored in Figure 2, for example, the system recommends urgent maintenance. It can also be seen that the software has taken vibration data into account for this purpose, as well as the dynamic maintenance interval to be monitored.

Comprehensible Results - Even Without AI Know-how

The difference to common methods lies less in the results of the forecasts than in the form of their presentation, which enables users without AI expertise to understand and evaluate the basis for decision-making.

Thus, users can confirm or reject the recommendations or adjust the sensitivity of the labels via sliders. From this feedback, a integrated learning algorithm in turn derives further patterns and learns continuously via an integrated machine learning process.

Step by Step to a Predictive Asset Management Strategy

Anyone who operates machinery or asset parks must find a good balance between the highest possible availability and the lowest possible maintenance costs.

Balance between the highest possible availability and the lowest possible maintenance costs can be achieved by holistic and consolidated asset management.

Optimized, the relevant interactions can be managed by using artificial intelligence methods, especially if these show those people in charge of the process recommendations for action, the evaluation of which does not require any knowledge of AI.

Field-proven Solutions for Optimized Maintenance and Service Management

Software solutions for optimized maintenance and repair management are also practical if, in addition to suitable scaling options, such solutions can also map the entire process from maintenance recommendations and concrete planning of maintenance operations to continuous monitoring of the processes (Figure 3, left), e.g., by means of messages on the processing status of maintenance and repair operations, and are therefore suitable for both predictive maintenance and predictive asset management scenarios.

In the example, this is done by taking into account other influencing factors that are processed using the same systematics and the principle of Qualitative Labeling. Likewise, the learning logic that can be used in the background can learn interactions and systematics at a high scaling level. Consequently, the only change concerns scaling, e.g., with regard to the use of databases and further maintenance management tools such as PSIcommand.