The requirements for machine maintenance have changed significantly in recent years. Instead of following fixed maintenance intervals, industrial groups are increasingly focusing on artificial intelligence (AI) to detect faults before they occur. This enables them to minimize or avoid breakdowns and unplanned downtime, thereby reducing maintenance costs.

This is made possible by predictive maintenance. AI can be used to continuously record and analyze process and production data from machines and systems and identify deviations outside the tolerance limit. Based on the collected data, the system calculates maintenance information, such as the ideal time to replace a wearing part in order to prevent long-term machine failure.

Challenge: Mastering data and making it usable

During predictive maintenance, a huge amount of data is produced and recorded every day. Sensors installed directly on the machines and systems continuously measure defined parameters such as temperature, humidity, pressure, etc. At the same time, intelligent machines document information about the environment. At the same time, intelligent machines document information about the environment, for example outside temperature, vibrations or humidity. In this way, deviations outside the tolerance limit can be detected quickly and faults rectified at an early stage, even before a standstill or failure occurs. Only when combined with other data available in the company, such as maintenance plans, construction drawings, orders and invoices, does this collected information lead to in-depth insights and thus to an enormous advantage in a highly competitive market.

In practice, however, it is usually the case that the maintenance teams work with completely different applications and programs than the employees in purchasing, accounting or quality assurance. The available information is therefore distributed across different applications and each employee only works with a fraction of the theoretically usable knowledge. What is missing is the ability to map an overall view of a specific topic.

Figure 1: Interactive Exploded View of an Insight Engine.

© Mindbreeze

In order to access this overall view, companies are increasingly using so-called insight engines. They combine technologies from classic enterprise search solutions with artificial intelligence and advanced speech recognition methods to intelligently link, analyze and process company data from a wide variety of source systems. If users are looking for a specific topic such as a component, the Insight Engine draws from the entire data pool. It takes into account all unstructured and structured data - from production data to maintenance logs, plans, documentation and expert opinions - extracts the required information and provides it enriched with context-specific additional information. Information on order frequencies, quality or experts on a searched topic or the corresponding component is thus also available in a clearly structured form. In the form of an 'Inter-active Exploded View', this detailed information can be called up by mouseover without the user having to start another search. This provides the opportunity to identify potential for improvement at a glance and, if necessary, to adapt individual production processes or even entire business processes. If a component needs to be replaced due to a deviation outside the tolerance limit, this 360° view provides an overview of all relevant information on the manufacturer, supplier or quality, even though it is stored in different data sources(Fig. 1).

Difference through artificial intelligence

In contrast to conventional search engines, insight engines use artificial intelligence and speech recognition methods. Using machine learning and deep learning as well as relevance models and rules, they are able to extract the required information from the available data and link it according to its correlation.

Various speech recognition methods such as Natural Language Processing (NLP) and Natural Language Understanding (NLU) are used to learn linguistic subtleties, similarities and meanings of different sentences. While NLP ensures that the insight engine processes human language correctly, NLU ensures that the user's intentions are interpreted correctly. Despite the imprecision and ambiguity of human language, insight engines are able to correctly deduce the user's intention from the question and align the results accordingly. Users only receive the answers that are relevant to them, clearly presented in personalized dashboards.

Figure 2: AI-based relevance model calculated from user behavior.

© Mindbreeze

Machine learning can be used to identify specific patterns in data and derive predictive models. The Insight Engine uses various methods to do this. One of the best-known methods is the artificial neural network model (deep learning), which simulates a network of neurons based on the human brain. In addition, the system learns from user behavior and calculates a model (relevance model) that automatically prioritizes and proactively displays more relevant content(Figure 2). The specific and personal access authorizations of the users are always taken into account. The solution checks these directly at the individual data sources for each search query and restricts the results accordingly. Data for which a user has no authorization is therefore not available as a search hit. In this way, users receive their overall view (360° view) of the company's knowledge - within the scope of their respective authorizations.

Introduction of insight engines

Insight Engines can be integrated into the existing IT infrastructure and familiar working environment. Connectors are used to connect all data sources, i.e. the applications in which the data is stored. Depending on the provider, several hundred connectors are available 'out-of-the-box' for a wide variety of source systems. They are also available in different operating models.

One option is the on-premises appliance: Here, high-performance hardware with pre-installed software is integrated into the company's internal data center without any connection to the outside world. For companies that already manage a large amount of data in cloud solutions (SharePoint online, Office 365, Salesforce), a SaaS solution (Software as a Service) is an option. In this case, the appliance is located in the provider's cloud data centers.

The hybrid solution is a combination of both models: The appliance is used to analyze the data from the various applications in the in-house data center, while the Insight Engine indexes the data from cloud services directly from the cloud.

On-premises solutions are often the preferred option, especially when it comes to sensitive information. This flexibility reduces the workload for employees in the IT department to a minimum and the Insight Engine can be deployed quickly.

The author: Daniel Fallmann is the founder and CEO of Mindbreeze.

© Mindbreeze

Once the data sources are connected, the Insight Engine analyzes the content of the various files and documents, links them together, interprets correlations and summarizes them in an index. Only when interacting with a result, such as opening or editing a document, is the user forwarded to the relevant application.

The use of intelligent insight engines does not end with predictive maintenance: with capabilities in the area of intelligent data preparation, they represent an effective tool for the entire manufacturing company - from skills management to BOM management and quality management through to the areas of service and digital twins.