AI in the construction planning

... first steps to analyze content using AI.

Our use case in the construction planning

In construction planning we deal with the following three use cases: 

1. The review of service specifications: In this use case, the current bill of quantities should be checked on completeness for the client in order to predict possible amendments. The goal is to support cost stability and to reduce annoying testing work. 

2. The prediction of completion dates: Based on identified process sequences and their durations, options for project schedules will be created. The aim is to create easily a schedule that is based on empirical values and thus guarantees time stability. 

3. The prediction of deficiencies: This use case includes the prediction of deficiencies before the start of construction. The aim is to achieve early quality stability.

In all three use cases, each construction site – even each project participant – uses free text o document information in different ways. For example, when defects are recorded, a defect title and a defect description are usually documented. A direct comparison of the contents is therefore not possible. And a preparation of the data with intelligent methods is required.

Artificial intelligence for text processing

Using the example of deficiency prediction, a possible procedure for processing the information is explained below:

Since there is no standardized wording for deficiencies, different site managers, for example, name the same deficiency differently. An evaluation is therefore impossible.  

This is where Word Embeddings help us come into play. Words are represented by vectors in these procedures. One of the best known methods is word2vec. By vector representation, mathematical operations can be performed with words. Thus distances can be calculated and similar formulations can be recognized. With these Embeddings help us very good results can be achieved in Natural Language Processing (NLP) tasks such as Syntactic Parsing or Sentiment Analysis.  

With this method we reach the first important milestone in free text analysis. Individual words are classified and similarities are found. An evaluation of the data regarding deficiencies is thus made possible.  

The application of word2vec to a data set for defect prediction

But not only in this use case Word Embeddings help us: Even for schedules and amendments there are almost without exception free text fields. With our construction-specific word vectors, these can also be understood by the machine.

Here you will find a summary of the three use cases: