Predictive Maintenance Implementation: Why Data Collection Comes First

Key takeaways

A predictive maintenance implementation succeeds only when reliable, historicised and contextualised production data is in place first. At an aerospace and defence manufacturer running more than 70 CNC machining centres, AI/ML models could not predict downtime because 99% of stoppages happened in manual mode. The data revealed the real cause, human error, and targeted staff training significantly reduced downtime and improved overall KPIs.

A predictive maintenance implementation is only as good as the data behind it. At TT PSC, we believe that manufacturers can achieve the real goals of Industry 4.0 in manufacturing only when they rely on reliable data collected in real time. On that foundation, knowledge and awareness of what is happening on the shop floor can be built, and only then can the right improvement activities be identified. This project from the aerospace and defence sector shows why, and what it means for any digital manufacturing programme.

Why Predictive Maintenance Implementation Starts with Data, Not Algorithms

Predictive maintenance (PdM) uses machine condition data and machine learning to forecast failures before they stop production. In a smart factory, it sits on top of several layers: industrial connectivity that reads data from PLCs and CNC controllers, a historian that stores it, context from manufacturing operations management (MOM) and MES systems, and industrial analytics that turns it into decisions. If any of these layers is missing, the models have nothing trustworthy to learn from.

The case below shows what happens when a manufacturer aims straight for prediction, and why the most valuable outcome was the visibility that the data provided.

The Challenge: Predictive Maintenance for 70+ CNC Machining Centres

The aerospace and defence industry places particularly high demands on precision and reliability. Maintenance KPIs, failure prediction and machine condition monitoring are therefore crucial. Our client, a manufacturer operating more than 70 CNC machining centres, wanted to monitor its production process and optimise its metrics by implementing predictive maintenance. The main goal was to predict machine downtime, which until then had significantly disrupted the flow of production.

How the Predictive Maintenance Implementation Was Built

The project covered four building blocks of a connected factory. Each of them delivered value on its own, before any prediction was attempted.

Data Historicisation and Storage

Readings and calculations were collected and stored over time. This historical record made it possible to analyse trends, monitor equipment status and draw conclusions for the future.

Machine Connectivity and Integration with Production Systems

Using Industrial Internet of Things (IIoT) technology, the machines were connected and integrated with production systems. Machine performance analysis then made it possible to monitor machine status and production processes continuously and in real time. In a modern connected operations architecture, this layer is typically built with a connectivity platform such as Kepware and structured in a Unified Namespace, so that every system works on the same data.

Sensor-Based Machine Condition Monitoring

Additional sensors were installed to monitor equipment parameters that had previously been completely unavailable. They provided the data needed for analysis and enabled, for example, better monitoring of the production process.

Visualisation of Machine Status and KPIs

Dashboards visualised machine statuses and KPIs. This enabled ongoing monitoring and a rapid response to anomalies and alarm conditions, which minimised the occurrence of failures. The same data foundation also supports production monitoring with OEE.

Why the First AI and Machine Learning Models Fell Short

The initial results from artificial intelligence and machine learning (AI/ML) did not meet expectations. The AI/ML engine was unable to build adequate predictive models. The analysis showed why: 99% of the downtime occurred while the machines were in manual mode. These situations therefore could not be predicted by AI/ML algorithms.

The Real Root Cause: Human Error on the Shop Floor

The data exposed the main problem: downtime caused by human error, which the AI/ML engines struggled to predict. In response, the client provided additional training for its staff. This significantly reduced downtime and improved overall KPIs.

The project reached its business goal of reducing downtime. It did so through a people-focused action identified by production analytics, not through a predictive model.

Manufacturing Data Collection: The Foundation of Predictive Maintenance

The client came to the project interested above all in the capabilities of predictive maintenance. The real problem, however, turned out to be the lack of prior work on data: its verification and archiving. Without that work, there was no way of knowing whether a PdM solution would work at all.

The system TT PSC developed did more than monitor production data. It also provided analytics that helped identify the main problem, human error, which AI/ML could not predict. Solving that problem through staff training proved crucial.

Skilful data collection and storage is the foundation of effective production management and a key element in implementing advanced systems such as predictive maintenance. Without solid data, any analysis or predictive model becomes worthless. This is why data and AI initiatives in manufacturing should begin with connectivity, historisation and data quality rather than with model selection.

Context Turns Data into Information

Knowing that a CNC machine stopped is data. Knowing that it stopped in manual mode, during a specific order and shift, is information. In a connected factory, much of this context comes from MES and MOM systems. This is why predictive maintenance works best as part of a wider manufacturing operations management landscape rather than as a standalone tool.

Data Readiness Checklist for Predictive Maintenance

Before starting a predictive maintenance implementation, check whether your plant can answer yes to these questions:

• Connectivity: are all critical machines, including CNC controllers and PLCs, connected and delivering data in real time?
• History: is the data stored in a historian long enough to show trends, normal behaviour and past failures?
• Context: is each reading linked to the machine’s operating mode, order and shift, so that automatic and manual operation can be told apart?
• Quality: is the data verified, complete and archived consistently?
• Baseline: are downtime, maintenance and performance KPIs measured today, so that any improvement can be proven?
• People: do operators and maintenance teams understand how their actions affect machine data and downtime?

If several answers are no, start with data collection and production monitoring. These steps deliver value quickly and build the dataset that predictive models will need later. Our step-by-step roadmap for digital transformation in manufacturing shows how to sequence them.

Lessons for Your Predictive Maintenance Implementation

• Take small, careful steps and keep context in mind: it is impossible to achieve all ambitious goals at once. A realistic approach and phased implementation of new technologies are key.
• Start with data collection and awareness of production status: collecting data first gives you a complete picture of the actual state of production.
• Turn data into information: context and the analysis of historical data transform raw data into valuable information.
• Divide the work between people and algorithms: people should perform creative tasks, while algorithms handle precise and repetitive operations, which helps avoid human error.
• Combine technology with people: this predictive maintenance project at an aerospace and defence company showed that the technology, while powerful, needs the right context and collaboration with people to deliver the intended results.

Predictive Maintenance as Part of a Digital Manufacturing Strategy

A predictive maintenance implementation is not a standalone IT project. It is one use case within a wider digital manufacturing strategy that connects machines, people and systems across the smart factory. The same data foundation that enables prediction also powers real-time production monitoring, quality analytics and, over time, digital twins of assets and processes. Manufacturers that treat data collection as the first milestone, rather than a side task, build predictive capability on solid ground. For more examples, see our overview of Industry 4.0 use cases with real results.

Frequently Asked Questions