Predictive maintenance – the future of maintenance

Unplanned machine and plant downtime leads to costly stoppages and productivity losses across all industries. Predictive maintenance (PdM) offers a solution. This guide explains how PdM works and what technological requirements are necessary for its successful implementation.

Unplanned machine and plant downtime leads to costly stoppages and productivity losses across all industries. Predictive maintenance (PdM) offers a solution.

By continuously monitoring machines and plants, it is possible to predict when maintenance work is required before a breakdown occurs and, in the worst case, production comes to a standstill. But how does this method work and what is needed to implement it in a company?

This guide explains how predictive maintenance works and what technological requirements are necessary for its successful implementation. It also discusses the opportunities and challenges of this maintenance strategy, with a particular focus on how companies can increase their efficiency by optimising their maintenance processes.

The idea of predictive maintenance is not new. The first concepts were developed in the aviation industry back in the 1940s to prevent aircraft failures. However, it was not until the advent of Industry 4.0 and the Internet of Things (IoT) that predictive maintenance became a core topic in maintenance.

Traditionally, a distinction is made between reactive and preventive maintenance in the field of maintenance. In reactive maintenance, machines are only repaired once they have already broken down. Preventive maintenance, on the other hand, is carried out at fixed intervals, regardless of the actual condition of the system. However, both approaches have disadvantages.

Reactive maintenance can lead to higher costs in the long term, as it often results in premature wear of components and a shorter service life for machines.

In contrast, preventive maintenance requires the replacement of parts that may still be functional and the implementation of complex maintenance programmes. This approach is time-consuming and leads to regular interruptions in production processes.

Predictive maintenance goes one step further. Sensors and AI-supported data analysis are used to determine the optimal time for maintenance measures. This combines the advantages of both approaches and minimises their disadvantages.

One example is the predictive maintenance of tools on turning and milling machines. During a machining process, the tool wears out and, at a certain point, vibrations or chatter can occur on the tool due to wear. These vibrations lead to further stress until the tool eventually breaks. Sensors can be used to detect wear-related vibrations and initiate a timely tool change before a failure occurs.

This prevents damage to the component being machined, such as torn drill holes or, in the worst case, damage to the machine due to flying parts.

Predictive maintenance combines machine data and artificial intelligence (AI) to determine the optimal time for maintenance measures. This approach is underpinned by sensors, IoT devices and advanced analytics software that continuously collect data on the condition of machines and systems.

The process can be divided into four interdependent steps. Each of these steps requires specific technologies and skills, which are examined in more detail below.

The effectiveness of predictive maintenance is based on the interaction of various technologies. While sensor technology, the Internet of Things (IoT) and machine learning have already emerged as core components of how it works, technological advances in these areas have further increased the performance of PdM systems.

In the field of machine learning, deep learning models have proven to be groundbreaking for PdM. Convolutional neural networks (CNNs) are particularly characterised by their ability to recognise complex patterns in multidimensional data.

In practice, CNNs are used, for example, to identify anomalies in the vibration spectra of rotating machines. This allows even subtle changes in vibration patterns that indicate incipient bearing damage to be detected at an early stage.

However, these advanced technologies are just the tip of the iceberg. Fully exploiting the potential of PdM requires additional technologies for processing and analysing the enormous amounts of data generated, as well as for seamless condition monitoring. Two further essential technological foundations are therefore considered below.

A major challenge in implementing predictive maintenance is the integration of different data sources. To obtain a comprehensive picture of the plant status, data from a wide variety of systems and sources must be consolidated and harmonised.

The complexity of this task is further increased by the variety of data formats: from simple numerical values to text files and log files to complex structured data such as JSON or XML – the heterogeneity of the data places considerable demands on the integration solution.

Industrial Internet of Things (IIoT) gateways offer a powerful approach to overcoming this challenge. These specialised devices act as a bridge between the physical world of machines and sensors and the digital world of data analysis and decision-making.

IoT gateways enable seamless communication between machines, sensors, ERP systems and external data sources such as transport and logistics data. Their core function is to translate and standardise the various data streams, regardless of the underlying systems and protocols.

A key advantage of industrial IoT gateways is their ability to process and normalise data directly at the source. This means that the different data formats can be converted into a uniform format at the point of origin.

For example, proprietary binary formats or unstructured text files can be converted into defined structures such as JSON or XML, or unstructured log files can be converted into structured data records. This pre-processing not only reduces the amount of data that needs to be transferred to central systems, but also makes subsequent analysis much easier.

Condition monitoring and predictive maintenance are closely related, but differ in their focus and objectives. Condition monitoring focuses on the continuous real-time monitoring of machine condition and the detection of current problems and anomalies.

Advanced sensor technologies, in particular MEMS sensors (Micro-Electro-Mechanical Systems), have significantly improved data acquisition in this area. These miniaturised sensors enable the simultaneous measurement of multiple parameters such as vibrations, temperatures and accelerations in a highly integrated format. The result is significantly higher granularity and precision in condition monitoring, which detect and analyse even the smallest changes in machine condition.

In combination with IIoT platforms, these sensors form the basis for comprehensive real-time monitoring. Central systems then receive and process the data streams using advanced analysis methods such as frequency analysis or wavelet transformations. These techniques make it possible to decode complex signal patterns and identify anomalies at an early stage.

In summary, condition monitoring forms the foundation for predictive maintenance by providing the data basis for further analysis and prediction models.

MaibornWolff collaborated with a leading manufacturer of safety technology to develop a live monitoring platform for networked gas measuring devices. The project aimed to ensure consistent safety levels in industrial plants with gas supply lines by networking heterogeneous flow meters on a central platform.

This project illustrates an important trend in industry: many companies are initially opting to implement condition monitoring as a precursor to predictive maintenance.

The reason for this is the step-by-step approach to the digitalisation of maintenance. Condition monitoring enables companies to first build up a solid database and gain experience with real-time monitoring before moving on to more complex predictive analyses.

Predictive maintenance enables companies in various industries to move away from fixed maintenance intervals and reactive repairs to a data-driven, proactive maintenance strategy.

Combining real-time data from condition monitoring with advanced analysis methods opens up new possibilities for optimised plant operation and more efficient maintenance. This results in a number of concrete benefits:

• Reduction of unplanned downtime and optimisation of maintenance intervals
• Increased cost and operational efficiency
• Extension of plant service life
• Improved safety and compliance
• Optimisation of resource utilisation

A well-known example of the use of predictive maintenance is Siemens. The company uses this technology in various areas, including energy generation and gas turbine maintenance.

Siemens uses its Senseye Predictive Maintenance solution to automatically create behavioural models of machines and maintenance personnel. These models are designed to direct users' attention and expertise to the most urgent problems. The software analyses machine and maintenance data using machine learning algorithms and presents users with notifications within static, self-contained cases.

Siemens recently expanded its predictive maintenance solution by integrating generative AI. This new functionality is designed to make predictive maintenance more conversational and intuitive. It enables an interactive dialogue between the user, the AI and maintenance experts, which should make the decision-making process more efficient and effective.

Although the implementation of predictive maintenance promises significant advantages, it also presents companies with complex challenges. In the following, we therefore look at two areas that require particular attention during implementation: data quality and system integration.

Before the collected data can be used for predictive analytics, it must be carefully cleaned and prepared. Raw data collected by sensors and machines often contains errors, gaps or inconsistencies. These problems can have various causes:

• Measurement errors: Sensors provide inaccurate values due to calibration problems, environmental influences or wear and tear.
• Communication problems: Data transmission errors lead to missing or corrupted data points.
• Incomplete data: System failures or scheduled maintenance work lead to gaps in the data sets.
• Format inconsistencies: Data from different sources is available in different formats or units of measurement.

Another challenge for data quality in predictive maintenance systems is dealing with anomalies and outliers. Anomalies are data points that deviate significantly from expected behaviour or patterns.

They can arise in various ways, e.g. due to sensor errors or external influences such as sudden temperature changes. Misinterpreted anomalies can lead to critical machine problems being overlooked and unnecessary maintenance measures being carried out.

The integration of predictive maintenance into existing production environments presents companies with technical and organisational challenges. Many production environments use machines that have been providing reliable service for decades.

However, these older systems do not have the necessary modern sensor technology. They also lack integrated communication interfaces for data transmission and often cannot connect to modern networks.

The heterogeneity of machine fleets further exacerbates this problem. In a typical production environment, machines from different generations, manufacturers and technologies can be found side by side. This diversity leads to the following problems:

• Different data formats and structures
• Incompatible communication protocols
• Varying control systems and user interfaces
• Inconsistent maintenance and servicing requirements
• Diverging life cycle phases and spare parts supply

Integrating all these different systems into a uniform predictive maintenance solution often requires complex adjustments and interface developments.

In addition, heterogeneity makes it difficult to develop uniform analysis methods and prediction models, as each machine type has its own specific operating parameters and wear patterns. But even if a network connection is possible, incompatible communication protocols can complicate data transmission.

Another challenge is the implementation of the predictive maintenance model, in which anomalies and predictions are to be detected automatically on the basis of the data available. This requires a combination of data and expert knowledge, which is converted into algorithms.

To improve prediction accuracy and deal with incomplete or noisy data, the algorithms are continuously trained and optimised using new data. Developing effective models therefore requires close collaboration between data experts, domain experts and IT specialists, who contribute their expertise and translate it into technically feasible solutions.

The successful implementation of predictive maintenance in companies requires structured and creative solutions, especially when it comes to complex challenges such as cleaning up faulty data.

In order to effectively overcome technical and organisational hurdles, it is crucial to lay a solid foundation in the initial phase of the project.

MaibornWolff offers comprehensive consulting services that begin with a detailed analysis of your individual requirements and accompany you through all project phases. Our approach to holistic solutions comprises the following core elements:

At the outset, we conduct a comprehensive needs analysis that evaluates existing maintenance strategies and identifies specific goals and KPIs that you want to achieve with predictive maintenance.

An IIoT solution assessment determines the current maturity level of your maintenance processes, among other things, and identifies potential for optimisation within your IT organisation. We also take organisational factors such as employee skills and existing data management practices into account.

When developing predictive maintenance solutions, we place great emphasis on scalability to ensure efficient transfer to different plants or production lines. This enables cost-efficient implementation throughout your entire company.

At the outset, we work with you to identify specific challenges and potential, such as the heterogeneity of your machine park or the effects and signs of failures. Through a thorough analysis of the problem, we create the conditions for developing a customised approach to implementing predictive maintenance in your company.

Based on the insights gained, we develop tailor-made concepts that take both technical and economic aspects into account.

Our experts support you in creating a detailed ROI calculation to evaluate the benefits of the predictive maintenance solution. This enables you to make informed decisions and optimally leverage the advantages of predictive maintenance for your company.

When developing predictive maintenance, analysing the existing data base is a crucial step. Together with you, we determine whether all the data required for implementing predictive maintenance is already available or whether additional sensors need to be integrated and further data sources tapped.

Depending on requirements and circumstances, predictive maintenance can be implemented on individual machines or scaled across the entire plant. For implementation on a single machine, powerful industrial PCs (IPCs) can be used to run the machine learning algorithms directly. This approach enables targeted monitoring and optimisation of critical machines without the need for a complete IIoT landscape.

For plant-wide implementation, we support you in developing a comprehensive data strategy. Together, we design measures to optimise your data infrastructure and ensure data availability and quality.

A key element here is the implementation of a unified namespace, which ensures a consistent and standardised data structure across all systems. Modern approaches such as Data Mesh can be used to organise data management in a decentralised manner and transfer responsibility for their own data products to your teams.

Regardless of the scale, we place particular emphasis on future-proofing when identifying and installing the necessary sensors and data acquisition systems. We support you in setting up and integrating IoT data platforms that fit seamlessly into your existing IT systems while remaining flexible enough to integrate future technologies.

When implementing predictive maintenance, MaibornWolff relies on a structured and iterative process based on a deep understanding of the complex processes and challenges in your company.

At the beginning of the implementation process, we support you in selecting the right ML platform that meets the requirements of your predictive maintenance application. We focus on factors such as scalability, integration capability and user-friendliness to ensure seamless integration into your existing IT landscape.

We then focus on training the predictive maintenance models. This involves the use of modern machine learning methods that learn to predict potential failures and maintenance requirements based on historical data and real-time information. Through extensive testing and an initial evaluation, we ensure that the trained models meet the desired requirements and deliver reliable results.

To validate the practical suitability of the developed solution, we first start with a proof of concept (PoC). This allows us to test the models under real conditions and gather valuable feedback. Based on the findings from the PoC, we make any necessary adjustments to further optimise the performance and reliability of the solution.

After successful validation in the PoC, the predictive maintenance solution is transferred to live operation. During operation, we continuously monitor the predictive accuracy of the models and compare the results with reality.

If, for example, unforeseen failures occur or false alarms accumulate, we make targeted adjustments to the models to improve prediction accuracy. This iterative process of evaluation and optimisation ensures that predictive maintenance always remains up to date and delivers reliable results.

The successful implementation of predictive maintenance requires a sustainable training and development programme for employees. This aims not only to impart technical knowledge, but also to establish a culture of data-driven decision-making.

The training courses cover technical aspects such as the use of sensors and analysis software, but also the interpretation of data and its translation into concrete recommendations for action. Employees learn how predictive maintenance is integrated into existing processes and how they can contribute to the continuous improvement of the system. The goal is to enable your company to use predictive maintenance independently and effectively after completing our consulting services.

At MaibornWolff, we take an agile approach in which optimising the implementation strategy always goes hand in hand with its further development. By regularly reviewing and refining the prediction models, we ensure that forecast accuracy is continuously improved.

We also take changing production conditions into account to maintain the relevance and effectiveness of the system. This iterative process enables us to respond flexibly to new challenges and achieve a fast return on investment (ROI).

Another focus of our optimisation process is the integration of new technologies. We continuously evaluate innovative approaches in the fields of data analysis, machine learning and IoT technologies in order to further exploit the potential of the predictive maintenance system. This can include, for example, the implementation of advanced deep learning algorithms or the use of edge computing for faster data processing.

Once the predictive maintenance solution has been successfully implemented on a plant and has proven its effectiveness, we support you in rolling it out to other machines and production lines.

Thanks to the scalable structures created beforehand, the system can be efficiently transferred to other areas. In doing so, we adapt the solution to the specific requirements of the respective machines and take into account possible differences in data sources and process flows.

MaibornWolff supported a leading provider of sensors, controllers and systems for industrial automation in the development of an advanced remote access solution. This solution not only enables remote maintenance of plants and machines, but also forms the basis for the implementation of predictive maintenance. Remote access allows important machine data and sensor measurements to be recorded and analysed in real time in order to identify potential failures at an early stage and initiate proactive maintenance measures.

Challenge

The challenge was to develop a remote access solution that stands out from competing products in terms of ease of use and integrates seamlessly into the customer's existing IoT platform. Complex technical requirements had to be met while ensuring an excellent user experience.

Solution

Working closely with the customer, we developed a full stack cloud application and two RUST-based clients (Edge and Windows). The development process included:

• An upstream concept phase to define a minimum viable product (MVP)
• Implementation in a mixed team of customer developers and MaibornWolff experts
• Focus on ease of use and sophisticated error handling for an optimal user experience
• Close collaboration with the customer's product owner, UX/UI team and platform experts

Result

The project resulted in a remote access solution that is fully integrated into the customer's platform and was presented at the Hannover Messe 2024 trade fair. The solution offers:

• Efficient remote access to plants and machines
• Predictive maintenance functions for forecasting maintenance requirements
• Intuitive operation and improved error diagnosis
• Optimisation of maintenance processes and minimisation of downtime

The successful collaboration between our customer and MaibornWolff demonstrates how specialised technical expertise and a user-centric approach can lead to innovative solutions in the field of industrial IoT and predictive maintenance.

Predictive maintenance offers future-oriented companies enormous potential for optimising maintenance processes and increasing competitiveness.

As an important use case of smart factory concepts, predictive maintenance paves the way for intelligent and connected production. The insights gained and the creation of a future-proof data infrastructure serve as a valuable foundation for future projects and enable more efficient implementation of further Industry 4.0 initiatives.

MaibornWolff supports you in the consultation and implementation of this technology, from conception and pilot projects to company-wide rollout. Our interdisciplinary team has extensive experience in research and development as well as in the integration of predictive maintenance into existing infrastructures. We accompany you on the path to customised solutions and work with you to expand the future viability of your company.