Tata Innovista is a platform that recognizes innovations across various companies within
Group. Tata iQ, in partnership with Tata Steel, got this prestigious award in the Core
Innovations category on 11th September 2020 a novel project aimed to reduce Sticker Breakouts in the LD shop,
leveraging Machine Learning techniques.
Sticker breakout is a common hazardous phenomenon that plaques the LD continuous casters in Steel manufacturing plants. A sticker is a rupture in the solidifying steel shell when it is passing through the mold to get cast. If unheeded, it can lead to a breakout – a potentially hazardous outcome. To detect formation of stickers within the mold, its wall is embedded with a series of thermocouples to register temperature signals within the mold. Sticker forming causes the temperature signal to fluctuate with a particular signature – captured by thermocouples within its vicinity. If a sticker alarm is triggered, the casting process is slowed down to allow the sticker/rupture in shell sufficient time to heal. While rule based logic aimed to detect the temperature fluctuation signatures associated with stickers can be successful in eliminating breakouts largely, however, such logic end up with a high false alarm rate, causing production to be affected substantially.
To counter this problem, and make the sticker detection mechanism more pointed and accurate, Tata iQ developed a Machine Learning model leveraging Boosting techniques and a lot of innovative Feature Engineering. The model was successful in reducing the false positive rate substantially and at the same time eliminating false negative rate (undetected sticker leading to a breakout). Such a balanced model has been effective in helping augment the production substantially through the last couple of years since its deployment. The solution is scalable and is thus being evaluated to be implemented in other casters across Tata Steel. A paper on this solution – jointly written between Tata iQ and Tata Steel - was accepted for presentation in the fifth International Conference on Advances in Solidification Processes (ICASP5) and the fifth International Symposium on Cutting Edge of Computer Simulation of Solidification, Casting and Refining (CSSCR5) - held as a joint event in Salzburg, Austria in June 2019.
Tata AIG have been awarded at The Economic Times Digital Warrior Summit &
scheduled on March 9, 2021.
Best Disruptive Deployment – AI Based Retention Approach for Private Cars
Private Car insurance makes up significant part of Tata AIG book and retention rates were low. An AI/ ML powered retention strategy was rolled out, which provided a lift of 5% on NOP monthly. A systematic approach was applied for monthly Private car retention base by scoring every customer on the model & defining strategy based on their behaviour. Based on the model output customers were segmented & retention approach was implemented accordingly.
Tata iQ has developed a thermal coal price forecasting solution for Tata Power – to
prices of specific indices of thermal coal – both in the short-term (1 to 3 months) and
to 12 months) horizon. Leveraging this solution to optimally time the buying of coal
also negotiate better, Tata Power had a better buying strategy and was thus able to cut down
substantially on the annul coal-buying spend. In recognition of the effectiveness of the
forecasting solution to help augment their coal-buying strategy, Tata Power awarded Tata iQ
Value Creation award on February 2021 for developing the solution.
Commodity price forecasting is tricky – so many different factors are usually simultaneously in play affecting prices of commodities that it is difficult to garner all of them and aggregate their effect on prices in an objective manner. The production, consumption, trade flows, inventory levels, regulatory policies, market sentiments, macro-economic factors – all can affect commodity prices, to varying degrees which varies across time. Various data sources were leveraged to procure and collate all the diverse but relevant data elements that can affect prices of different thermal coal indices. The next important hurdle was to establish causality – which of the different data elements affect future coal prices, when and by what amount. Different statistical and machine learning techniques were employed to categorically assess the different data elements and their causal effect on the price of each coal index. Appropriate time series forecasting models were thus developed – uniting all the different data elements to predict how future coal prices of specific thermal indices might behave in the future.
The entire solution is hosted on the cloud, so that it can be accessed by anybody anytime from anywhere. However, this required a thorough automation of the entire process – from data download to analysis and identification of the set of variables most important as of today and then use the shortlisted suite to develop a forecasting model that is best in terms of accuracy and consistency of performance. A highly efficient dashboard to house the entire automated solution has ensured that Tata Power is able to use the solution conveniently, have access to different news articles and insights affecting prices and of course, a self-learning and evolving intelligence sitting at the centre forecasting prices and at the same time continuously learning how to do it better.
Tata iQ received the best paper award in ANQ Congress 2020, which was held in South Korea.
Network for Quality (ANQ) consisting of all non-profit organizations in Asia that seek to
quality of human life by contributing to the progress of science and technology; and to the
development of industry through promotional activities for the research and development of
theory, methodology and application in the field of quality and quality management.”
140 research papers were selected from various Asian Countries and our paper was selected as one of the best papers. The title of the paper was "Reducing idle freight pay out and carbon footprint through application of operations research at Tata Steel India".
This paper talked about the implementation of 3 projects which helped TSL to reduce idle freight through better utilization of vehicle/ rake space for FG transportation. There were 2 part to this project, the 1st was developing the core optimization model which recommends how to stack FG in a vehicle and the 2nd part was a 3D visualization of FG loading in a Vehicle. 3D visualization give confidence to the user about authenticity of the recommendation.