AI & Deep Learning in MEP - Buildings Services design

AI and deep learning - The new trend

Progress of new digital technologies is something that all people experience either directly or through indirect ways in their daily life. Applications like Uber have democratize the world of on demand and luxury transportation and other like Instagram or Ytube give unlimited opportunities for new artists and performers to share their work with the whole world. All these at their fingertips and the comfort of their home.

Further to that more smart applications like the Google search engine, on-line adds and even more the self driving cars use technologies which are in the field of artificial intelligence (AI) and run under sophisticated algorithms that have been developed using machine learning techniques (ML). Machine learning is by the definition of Arthur Samuel (1959);

‘Field of studies that gives computers the ability to learn without being explicity programmed’

In simple terms this means that computers run advanced algorithms which are fed by significant amounts of data (inputs and labeled outputs), using them in a way to identify patterns within the data structures. At the end the algorithm creates a non-linear model which is trained to fit all these data without any programming intervention by humans. Given the machine a new set of input data the computer uses previous learned models to predict the requested output based on its experience.

Neural networks are machine learning algorithms for modelling of complex non linear systems like image vision or self driving cars. The more complex the machine learning problem becomes the deeper the neural network in terms of the number of layers required, to perform calculations on the input data.

This is called deep learning and has boost modern systems to the field of autonomous and automated operation. Computers become more intelligent and based on input data can make decisions. They not just apply certain ‘If-then’ conditions programmed by the users. The computers are trained on decision making process from a batch of previous similar examples (training set) and become more intelligent providing accurate decisions using all the knowledge and experience gained.

Traditional business model for a BS design team

Traditional business working models for most of the engineering firms, that provide building services design and consulting, is to build large teams of engineers from several backgrounds (mechanical, electrical, life safety, controls etc.) and various experience levels. The project is usually broken down into separate packages and distributed through small teams of disciplines. Graduate or low level engineers perform all initial time-consuming calculations in order to provide input data to the more senior engineers which are responsible for the development of the concept design and systems selection. This process can go through several rounds of trials and optimization efforts based on the project complexity and team experience. Use of computers in the process is only to perform repetitive calculations and use of CAD software for drawings.

Common issues appear within this process are:

• The team does not use data from similar past projects because of luck of updated and easy to access enterprise data bases.
• Limited or inefficient quality control adds risks for data calculated from the more graduate members of the team.
• All design steps are time consuming and demand large number of resources which affect the fast decision making on the early stages of the projects, critical for systems concept selection and investment decisions.

Digital transformation is a way forward to provide solution on some of these working problems. Using new technologies like machine learning can transform the role of engineers within the team structure. Implementing computers within the design process in intelligent ways through ML or even deep learning can make management of knowledge and past projects experience fast and efficiently used.

New era and the collective intelligence into the working environment

Core of this transformation shall be the creation of a working environment with a new form of collective intelligence. By the definition of professor Tom Malone (MIT) collective intelligence is:

‘Groups of individuals acting collectively in ways that seem intelligent’

In the case of a design team transformed for a new digital era the group is a combination of engineers and computers all linked together in intelligent ways to exchange information and knowledge, so to accomplish design tasks.

Tasks for computers shall be scheduled in ways that make use of machine learning algorithms in order to process vast amounts of history data and predict design values and elements fast and accurately.

Engineers shall spend more time on the evaluation of these outputs from the computers, use critical thinking, interact with other teams for the interfaces and develop the details of design.

Implication of a strategy like this in the working structure of a building services design firm has to go through several stages of transformation which is not part of this blog. Instead some examples of the way that intelligent computer applications can be created gives a feeling of the way ML and deep learning is implemented.

Applications of deep learning into the buildings design life-cycle

The life cycle of a building or infrastructure project design has incremental and thoroughly defined stages or phases. Either if the development is related to a new or existing project, the design procedure shall follow international accepted design phases. These phases allow multi-disciplinary design teams to search and collect project needs, requirements and specifications, identify interfaces and proceed with initial ideas that will form the concept design. Adding engineering knowledge and developing further the project produce the preliminary design. Finally performing detailed coordination and individual components and systems design ends the life-cycle into the detail design phase.

The concept phase is a one of the most critical stages during the development of a project design as a potential investment. For most of the investors it is the stage where the size of assets will provide them with an initial assessment of the project scale and financial impact. As that engineering firms have to consult their clients with a conceptual design that will predict the size of the investment in the most valuable way.

Consultants should develop tools which allow the prediction of all these elements that will affect the investment. Using machine learning and deep learning technology such tools can take the form of ML or Neural Network models that will use as inputs certain feature of the building design and provide predictions of required outputs like the expected power demand, cooling capacity or number of lifts just to name a few. Examples of such applications could be the following:

·             Using input features like building occupancy, location, climate conditions, indoor temperatures, expected system efficiency, number of cooling zones and other HVAC requirements, the model shall be able to predict the cooling and thermal capacity with accuracy.

·             Another similar model can be fed with input features from the electrical team, like building height, area, occupancy type, lighting technology, control system presence, hvac type and other inputs depending on the project usage and predict the total power demand.

The above are two high level descriptions of neural networks that could provide the team immediately with values that allow initial sizing of critical and high cost equipment like substations and cooling / heating plants.

Similar models can be trained by design teams based on their area of expertise and projects needs. Applications are numerous and prediction of design and engineering elements can also include the preliminary or detail design phases.

The magic at the back of this approach is that neural networks are models which provide immediate and accurate predictions. In order to achieve that require previous training with data from other similar projects, probably done by the team in the past, which are used as validated references. This process can be broken down into the following steps:

Output prediction	Selection of a valuable output design element that is important for the design development like cooling capacity or power demand.
Input features	Selection of all those design parameters that are strictly related and affect the expected output value.
Training Data	Collection of input / output data from similar past projects that will be used to train and test the model.
Model construction and training	Building of algorithms, selection of model parameters and training of the model.

Depending on the specialization area, design methods, working structure and experience of engineering firms, deep learning can have several applications and improve significantly the quality and efficiency of the design process.

Initial applications may seem small and without impact on the business but as long as data mining becomes more systematic and engineering teams think of different areas of applications, the results of digital transformation can become remarkably large. We are in the frontier of a new era and digital strategy for engineering consulting firms is a priority.



I hope that you find this post interesting and educative.

• If you did like it please share it through the social so that more people can have access to it.
• If you have any questions or would like to discuss any special case, please leave your comments below. I will be happy to answer!