Back to blog overview

How labelling helps you kickstart your predictive maintenance projects

by Tara van de Lagemaat, Account Executive Industry 4.0 on Jan 23

Labelling helps you kickstart your predictive maintenance projects. You may not know it, but it is highly likely that you have been labelling data already. The image below shows data labelling for the image recognition algorithms of Google. Google uses out input to create useful data that their algorithms can train on so they can eventually recognise certain objects on their own.

 

Likewise, labelling can also create the useful data that’s required for predictive maintenance algorithms. For image recognition algorithms, labelling entails marking images with a specific object. In the field of predictive maintenance, labelling is the act of marking relevant fault-related patterns in available machine data.

 

 

By creating valuable and useful machine data, you essentially tackle two major challenges in the industry, namely: the aging workforce and the knowledge drain. This is because the process of labelling allows you to store valuable machine knowledge and to keep this specific knowledge in the company.

 

To help companies kickstart their predictive maintenance projects, Widget Brain has developed a way to easily inspect and label relevant machine faults in the data. Once the faults are labelled and stored and our fault diagnosis algorithm is trained, it can automatically detect the patterns on new incoming data and tell you when your machines require attention. This helps you to take appropriate action when it is needed. So how does it work?

 

The labelling process in 5 easy steps

With Widget Brain’s tooling, you can label, store and detect relevant patterns. To explain what that looks like, we’ll go through the following steps:

  1. Connect a data source and label your first pattern, potentially using an event log as guidance.
  2. Automatically store your labelled patterns into a central knowledge base, that can be downloaded at any time.
  3. Train the self-learning algorithms on the patterns.
  4. Detect stored patterns in new data and validate the output.
  5. Establish a connection with your asset management system.

1. Connecting and labelling

 

 

The first step is to make sure your data is available. We connect the tool to any database, such as SQL servers, Data lakes and more, through an API. It is also possible to upload a single file, for example, a data extract from a historian in a .csv, .xml or any other format to this tool.

 

A label can be created by selecting the timeframe of your pattern of interest. It can be valuable to use an event log to point you towards periods where relevant patterns may have occurred. You can upload your event log and automatically add patterns to the tool.

 

2. Your digital knowledge base of patterns

 

 

After labelling your patterns, they are stored in a central machine-specific database. This is the place where you can easily keep track of how many labels of each pattern exist. By clicking on the download button, you’ll get a .csv file with all your labelled patterns. You can use this information to educate new engineers, to compare it to the event log and to start experimenting with the data a little more yourself.

 

3. Training the Fault Diagnosis algorithm

 

 

After having your first pattern labelled or adding new labels, you can train or retrain our Fault Diagnosis algorithm by clicking on the (re)Train button. Within a few minutes, this algorithm will ‘learn’ to continuously match new patterns with the labelled patterns that you have specified and stored in your knowledge base. The result? You will get notifications whenever the Fault Diagnosis algorithm detects new patterns that match with existing labels.

 

4. Detecting and validate detected patterns

 

 

After training the fault diagnosis algorithm, you can select a new time period to test if your pattern of interest is detected. When it didn’t pass the test, your machine experts can always adjust labels to increase the algorithm’s accuracy. This input is used by the Fault Diagnosis algorithm to continuously improve its detections and make sure that the output stays relevant.

 

5. Integration with the asset management system

 

 

Now that you have validated your output, it is time to start using the Fault Diagnosis algorithm in real-time. When it detects a pattern that matches your labels and has a significant impact on your machine’s uptime, performance, safety, or compliance, you want to know about it straight away. Existing examples of such warnings are simple notifications to operators, but we also send full-fledge service job suggestions in EAM or CMMS systems for the maintenance manager. The appropriate action will then automatically be sent to the medium of your choice, for example in your email or as a service job in Maximo, Ultimo, SAP or any other system.

 

To conclude

Labelling is a crucial step in the implementation of predictive maintenance in your organisation. It is a requirement for getting predictive maintenance projects up and running. In addition, labelling is an easy and intuitive way of storing your knowledge digitally and to anticipate new market trends such as the knowledge drain, the shortage of technically skilled people and tightening compliance laws.

 

For more information about how our AI services have an impact on your business, visit: https://widgetbrain.com/predictive-maintenance/.