The TLP4Heat project is aimed at developing a digitization workflow that allows for utilizing maintenance information in district heating systems to develop optimal maintenance strategies.

To ensure efficient operation of the district heating network, the maintenance activities at the district heating transfer stations can be optimized by using predictive maintenance strategies. A necessary prerequisite for this is the digitization of maintenance information, such as service reports. These contain the key information for the development of error detection models in free texts. However, this information can only be utilized if it is available in a standardized and structured form. The primary goal of the project is to develop a digitization workflow for the standardization and structuring of maintenance information in district heating systems.

 

The project is interesting to:

Utility companies, district heating network operators, component manufacturers

Goals

  • Analyzing existing maintenance information and technical guidelines for maintenance activities in the district heating sector
  • Investigating the use of AI-assisted tagging for automated data annotation
  • Developing a conventional machine learning classifier to serve as the baseline for the assessment of more complex methods
  • Investigating the use of word embedding vectors for the classifier development based on neural networks

Methods

  • AI-assisted tagging methods are intended to be used for data preparation, in particular the application of the National Institute for Standards and Technology by the name of Nestor. This makes it possible to create a labeled dataset for the classifier development.
  • Classifiers will be developed using Sklearn and Keras, with the former being used for the development of random forests and the latter for the development of neural networks.
  • Word embedding vectors are planned to be used to improve the classification of the neural network. However, the word embedding vectors need to be first developed for use in the field of heating.

Detailed project description

In a first step, the existing maintenance data in the district heating sector is analyzed with the support of the Energy Efficiency Association for Heating (AGFW). By preparing the data, the available information is to be transferred to a general scheme to serve as a basis for the development of classifiers. Before preparation, there are no labels that can be used for training a classifier. These labels are assigned by preparing the data using AI-assisted tagging. The labels describe, for example, the component in question, the measure implemented or, in the ideal case, even the error cause.

The last step is the development and comparison of various classifiers that automatically assign the texts in the maintenance data to the labels. Initially, a relatively simple machine learning classifier is planned to be used to establish a baseline for the assessment of more complex methods. As possible enhancements, a neural network and one using word embedding vectors are to be tested. The developed classifiers can then be applied on other free texts to categorize them and enable obtaining structured and standardized data.

The structured maintenance information can be used by operators to establish KPIs, supporting them in making strategic decisions and optimizing their maintenance processes. In addition, the structured maintenance data serves as an important basis for the further development of machine learning methods for error detection, such as clustering. Another possible application is a support system for field technicians that facilitates the description of maintenance operations and ensures their uniform description. The trained classifier can also be used for this purpose.

Project schedule

  • Acquiring existing maintenance information
    • Acquiring existing data
    • Acquiring technical guidelines
  • Preparing the existing data
    • Transferring the information to a general schema
    • Labeling using AI-assisted tagging
  • Classifier development
    • Baseline: simple machine learning classifier
    • Possible enhancement: neural network, use of word embedding vectors with BERT

Project partners

Fraunhofer IEE:

  • Holger Dittmer (holger.dittmer@iee.fraunhofer.de)
  • Marc-Alexander Lutz (alexander.lutz@iee.fraunhofer.de)
  • Edison Guevara (edison.guevara@iee.fraunhofer.de)

AGFW Projekt GmbH (https://www.agfw.de/)

  • Research ideas and results can be presented and discussed in the association‚Äôs regulatory bodies. In the long term, the project results may be included in the guideline. In the spirit of maximum public visibility, the project findings can be distributed to the association‚Äôs members via AGFW to directly address a relevant target group.

Enercity Netz GmbH (https://www.enercity-netz.de/)

  • The district heating network operator from Hannover already promotes the digitalization of district heating transfer stations within the scope of the joint research project entitled ‚ÄúDigitalization of heat supply structures in a virtual thermal power station ‚Äď Use of information and communication structures for optimized and forecast-based control of heat generation and utilization systems ‚Äď SmartHeat (project number 03ET1673A)‚ÄĚ in collaboration with AGFW Projekt GmbH and Fraunhofer IEE. For the TLP4Heat project, the company provides O&M information from the operation of the district heating network.

Aalborg Forsyning (https://aalborgforsyning.dk/)

  • The district heating company from Denmark provides O&M information from the operation of the district heating network.

Danfoss (https://www.danfoss.com/de-de/)

  • Danfoss develops measurement and control systems as well as transfer stations for district heating systems, among other products.
Project period

January ‚Äď June 2022

Dr. Anna Kallert

Therm. Energiesystemtechnik

Fraunhofer IEE

+49 (0) 561 804-1876

Share this Spotlight with your network.

PDET - Incorporating the results of automated energy trading into science
GRAF-CoWi - Combination of wind power forecasts