Making a foray into AI for applications in the AEC industry
Status quo of Architecture, Engineering, and Construction industry
The Architecture, Engineering, and Construction (AEC) industry has recorded the lowest productivity rate over the years. It seems to have been stuck in a time wrap and left behind in the transformation saga. AEC’s growth has fallen behind that of other industries for quite a long time, and there is a $1.6 trillion chance to close the gap. The critical question we need to ask ourselves is why? Is it a matter of cost or fear for ambiguity in the usage of recent technologies?
AEC is potentially one of the most vital data generating industry. However, with the most minimal level of Datamation (mining data). The AEC has a substantial component of development which is “data”, and yes, here the data is in enormous quantity. Why? Useful data originates from all over the place, from design plans to construction processes, equipment and many more. BDO Global iterates that one of the prevalent problems of AEC is the lack (?) of ability to collect, analyse and utilise data. Building Information Modelling (BIM) is only a hint of something more substantial. Unfortunately, a lot of organisations are still hesitant to fully embrace BIM even though there has been a significant push for advancement recently.
What is Artificial Intelligence?
When most people hear Artificial Intelligence (AI), the first thing that pops up in their mind is robots, a thought of a complex system consisting of feats and logic. From SIRI to autonomous vehicles, AI is gradually gaining new grounds. Artificial Intelligence is a term for describing when a machine/system mimics human cognitive functions, like problem-solving, pattern recognition, and learning without being explicitly programmed for one task. AI focuses on learning, reasoning, and self-correction. AI does not just automate, rather, it analyses data and generates new solutions.
The fundamental point of Artificial Intelligence frameworks is to be able to find, what makes individuals increase their performance and efficiency over time. It causes machines or systems to learn from experience, conform to new information sources and perform human-like errands. Some of the technologies in Artificial Intelligence include machine learning, natural language processing, computer vision amongst many. All these technologies are branches/subsets of AI.
Technologies of Artificial Intelligence
Data Mining is a process used to extract usable data from a large set of raw data. It refers to the method of analysing and discovering patterns in large data sets. This process includes tasks such as cleaning, normalising, visualising and dividing data into training, cross-validation and test sets.
Computer Vision deals with the ability of systems to interpret and understand digital images and videos. The system can process these visuals in the same way the human vision does and then provide results. Problems of face recognition, handwriting recognition, Environment detection and motion analysis in robotics are a part of this area.
Knowledge Representation is responsible for representing information about the real world to enable AI systems to understand and utilise knowledge to solve complex real-world problems.
Natural language processing (NLP) is a branch of artificial intelligence that enables PCs to comprehend, decipher, and control human language (e.g., spell check). Organisations can figure out what clients are stating about an item by distinguishing and extricating data in sources from social media. This investigation can give a great deal of data about the client’s decisions and choices. Chatbots and speech recognition techniques are developed here.
Machine Learning (ML) is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns, and make decisions with minimal human intervention. It learns from past data to output results/predictions. The level accuracy depends on the data sets available. Major approaches can be classified as supervised, unsupervised, semi-supervised learning, depending on how the training set is labelled. Learning models such as regression, classification, support vector machines, neural networks, clustering, generative models are normally developed and integrated on this subset.
Machine gets a brain, so not only does it do repetitive tasks but does so with some thought applied into it. This can help us to explore challenging problems by simulating such complex environments, so we can focus on the core of the problem. At the same time, the machine intelligently iterates through various possibilities for us simultaneously. The application of artificial intelligence is endless, which can be applied to multiple industries.
AI will help add intelligence to the existing processes in the sector. Not compulsorily as an application but as integration with existing techniques. It will help us make informed decisions and avoid the avoidable loss. AI helps to reduce human effort and error. Programmed systems can perform repetitive tasks in new situations each time, in an intelligent way. Its ability to display a form of intelligence to decide without much human guidance.
Having to carry out repetitive and boring tasks can be tedious in nature. These are tasks that could easily be handled using AI algorithms. These systems can perform billions of tasks in seconds. Interestingly the output can be customised as per the case. AI has a wide range of benefits due to its ability to show a sign of intelligence by making choices based on input and feedback.
We at DiRoots are also employing some of the machine learning models in our projects, for instance, clustering methods that belong to unsupervised machine learning domain, are being applied on 3D point cloud data for structure detection. The two images on the right show the results of a Machine Learning model for DiRoots Point Cloud Automation (each colour shows different clusters, the image below).
AI can learn to play chess as it adapts through progressive learning algorithms. Intelligent, you would say? Another example is AlphaGo, a machine learning-based system that has beaten the best human Go players in the world. The applications of AI are so numerous that this article is not capable of containing.
Pitfalls and Challenges of Artificial Intelligence
As much as we have talked about the benefits of AI, it is necessary to be aware of its pitfalls. Some ventures fail because they start to use AI just for the hype. Before blindly implementing algorithms from the library, there must be a formulation of the problem and precise definition of parameters and proper understanding of what the algorithm aims to achieve in order to tackle the errors while debugging. Some vendors claim they understand AI but do they actually? There are lots of controversies as people and organisations do not have the right idea about AI implementations. AI is a promising field, but it takes time to setup. The vastness of AI, although an advantage can also be a pitfall when you are new.
The general challenges are:
- One of the significant challenges of AI is data availability. The more the available data is refined, the more precise predictions and outputs of AI systems are. Most times datasets are available but of low quality. It is crucial to map out a way of organising it into a structured manner to enhance result validity. It basically uses what you give it, garbage in garbage out.
- AI is usually costly to implement coupled with its limited number of available professionals in the field. These systems need to be continuously updated, and proper maintenance carried out on it. A lot of establishments lack the internal capacity to implement AI and mostly do outsourcing for both implementation and support, which is usually of the high cost and also of questionable on terms of quality.
- A major fear of AI mostly cited is unemployment. PwC predicts that 30% of jobs will be automated by 2030 and about 44% of workers (primarily workers with low-level education) because most of their task can be easily handled with a lower chance or error.
Technical challenges run from handling data and deciding the right features to consider and identifying then avoiding overfitting and underfitting problems.
- Effective implementation of algorithms requires some level of proficiency and understanding of areas of mathematics such as Multivariable Calculus, Statistics and Linear Algebra. Building that knowledge from scratch is a task that discourages many to get into the fields and many times not adequate to all due to its complexity.
- Poor choice of a library (performance of a single algorithm differs from library to library).
- Numerous research papers get published sharing novel techniques to tackle challenges, but these are rarely accessible for common masses due to its rigorous nature.
How can AI change and improve the AEC industry?
It is no news that attention has been drawn on AI and is rapidly on the increase. According to the Globenewswire, the global AI-in-construction market was valued at USD 312 million in 2017 and is expected to reach USD 3,161 million by 2024, growing at a CAGR of 38.14% between 2018 and 2024. PwC in 2018 carried out a study inferencing that AI-based automation will increase the GDP of the US to $15 trillion by 2030.
At the point when a structure is developed, the succession of building and designing assignments must be represented to keep these groups from venturing out of arrangement or clashing. Using generative design to help generate multiple design options for models, help identify and mitigate clashes in models using machine learning. Construction sites are likewise utilising AI to go through the pattern in a task. Pattern detection software can be sent to search for normal patterns in the task. The AI can extrapolate from that point what may occur if a license is deferred or an episode occurs and go through numerous situations rapidly. This determining activity can help guarantee plans are set up to manage sudden circumstances.
Mariusz Gorczyca, an R&D structural engineer at Kingspan, talks about the application of artificial neural networks in static structural analysis. He worked on the prediction of stress with AI using metamodels to predict aftereffects of inconspicuous structures; however, inside a similar class of structures. With AI, he was able to predict stresses in individual members by a variation of 10 to 20% from the actual values. This also showed that AI is highly dependent on the data available.
While the advantages of AI incorporate automation prompting efficiency, decrease in human blunders and progressively smoothed out procedures, the negative impacts of man-made consciousness ought not to be ignored.
What are the known applications of AI in our daily activities?
Apart from common examples like smart virtual assistants like SIRI, Google Assistant, or Cortana, Chatbots on different pages on the internet, mobile applications, recommender systems, handwriting recognition and many more.
What to expect from this series of articles?
AI is an extensive subject matter, and the content of this article would not be able to encompass its entirety. The upcoming article will expound on the various Machine learning models by building all the required mathematical pre-requisites and give an intuition for the working of algorithms.