Operational data can indicate that something is changing in a machine. To understand what that change means, you also need to know about past faults, the corrective actions taken, and their results. During their presentation at our “ Plantvision ” Technology Insight 2026 event, Robert Velén and Filip Abrahamsson described how maintenance can be improved through digital work methods. This article summarizes their discussion on facility information, mobile tools, and predictive maintenance, with one recurring point: technology requires reliable data from day-to-day operations.
The information needs to be available where the work is performed
When drawings, manuals, and maintenance history are hard to find, time is spent searching, printing, and moving around. Knowledge easily becomes dependent on those who know which binder or system to look in. Robert therefore emphasized the value of structured facility information, where the documentation is linked to the correct object. Anyone navigating to a pump should be able to find the information specific to that pump.
Mobile access makes it possible to use the information while on-site. Maintenance staff can work on work orders, follow checklists, and report faults directly at Location. Photos can provide a clearer description of the problem, and notes on drawings can capture changes that need to be incorporated back into the documentation after a corrective action.
Robert recommended tablets for reading documents and performing tasks over an extended period of time. Cell phones can work well for quickly reporting faults, but the screen size can be limiting when viewing drawings and manuals. He also noted that older versions of maintenance systems may not support the company’s own mobile solution. A separate mobile solution integrated with the maintenance system was presented as an alternative when the maintenance organization does not manage the upgrades itself.
A completed work order needs to explain what happened
Filip returned to an everyday example: a pump breaks down, a work order is created, and after the repair, it is marked as complete. The system shows that the work is finished, but if no one describes the fault or the corrective action, important information is missing for the next analysis.
“Fixed the pump” illustrated how a brief report can provide far too little information. What broke? What was done, and what was the result? If a leaking pipe is repaired without either the problem or the action taken being recorded, the incident remains in the memories of those involved, but not in the history that should be used to understand recurring problems.
The structure of the reporting is therefore important. The speakers called for a standardized approach to documenting events and actions, so that information from different people can be used together. This does not mean that every report must be worded identically, but the difference between a detailed description and a brief acknowledgment affects what can be learned from the material.
Machine learning requires both operational data and context
Predictive maintenance aims to anticipate when a problem is likely to arise, so that corrective action can be planned before the failure occurs. Filip described how machine learning can be used to identify patterns in operational data and link them to historical events. The focus was on a model trained for a specific process or group of machines.
The data can consist of temperatures, vibrations, and other signals from control systems or stored process history. Work orders and fault reports from the maintenance system provide information about what actually happened and what actions were taken. Data on downtime and scheduled maintenance can provide additional context. It must be possible to interpret the measured values in relation to how the equipment was used and maintained.
Filip used a hypothetical inventory error as an example of how the model might recognize a combination of changing trends and issue an advance warning. The example illustrated the principle, not a guaranteed warning time or accuracy. The model’s performance depends on the data it has been trained on. The lecture also emphasized that the results need to be followed up: new information about what is correct and what is incorrect can be used to train and implement a new version of the model.
Use existing data and choose the technology as needed
A recurring recommendation was to investigate what process data is already available before purchasing additional sensors. Temperature measurements and other operational data may already exist at the facility but may be difficult to access or stored in separate systems. Making them available for analysis can be an important step.
Supplementary IoT sensors can be useful when a relevant measurement is missing. A vibration sensor mounted on the outside of a motor was cited as an example of how additional information can be collected for analysis. The speakers also made it clear that alerts do not always require AI. Rules based on known thresholds and combinations of signals can provide support, while machine learning can search for correlations that are more difficult to identify manually.
Camera analysis raises questions about purpose and privacy
During the question-and-answer session, one participant raised concerns about the use of camera monitoring in production and worries that the footage might be used to scrutinize operators’ mistakes. The discussion focused on clarifying the purpose, limiting what the cameras capture, and engaging in dialogue with employees and union representatives.
The speakers also outlined clear guidelines for its use and an approach that focuses on improving the process without singling out individuals. The discussion did not provide a general answer on how camera analysis should be handled legally, but it showed that these issues need to be addressed when a solution is being developed.
The next step begins with an analysis of the current situation
The speakers’ key takeaway was to first understand where your own maintenance organization stands. Is there a comprehensive history? Are faults and corrective actions documented? What operational data is available? The answers provide a basis for selecting measures that can have an immediate impact and for planning what needs to be developed over time.
Summary of the Lecture
The digitization of maintenance encompasses both access to information in day-to-day operations and the ability to predict future failures. Structured facility information and mobile work methods can support the work at Location, while consistent reporting builds the history needed for analysis. Predictive maintenance using machine learning requires that operational data can be linked to failures and corrective actions taken. Therefore, start with the current situation and choose the next step based on the actual needs of your operation.
What happens at your company when a work order is completed? Does the information remain in the system, or does it stay with the person who did the work? We’d be happy to discuss your maintenance processes further and what you’d like to get a better handle on. Get in touch, and we’ll work together to identify the next step that fits where you are today.