MACHINE LEARNING TECHNOLOGY
We are a team of highly motivated software architects and mathematicians with diverse backgrounds – from machine learning, routing algorithms, and signal processing to compiler technology and theoretical mathematics. We combine our knowledge to advance
CeleraOne’s technology and analytic power every day. We understand steel manufacturing processes and approach challenges systematically. Our analyses leads to substantial results within limited amounts of time.
Deep Neural Networks
Deep Learning outperforms other technologies in complex tasks including computer vision and speech recognition. Applications range from automated translation to medical diagnoses and traffic sign detection.
Its strength - the great modeling power of Deep Neural Networks - comes at the price of intricate challenges in the construction of Deep Neural Models. Drawing from its broad experience covering multiple industry sectors, CeleraOne overcomes these problems by careful modeling of the customer-specific problem and adaption of pre-optimized architectures.
Complex processes canonically exhibit time series data. For non-trivial problems, this data is non-stationary. Heisenberg's uncertainty principle states a fundamental challenge of practical time-frequency analysis. Extraction of very precise frequency information leads to uncertainty in time domain, and vice versa.
CeleraOne applies a variety of multi-resolutional models to time series data that are tuned to maximum information extraction for the specific problem at hand.
Significant visualizations and understanding of industrial data goes hand in hand. In steel manufacturing, 1D, 2D, and 3D datasets appear with varying data type, including process variables and distributions of surface defects on cold rolled strips.
CeleraOne uses interactive, web-based tools for accessing Machine Learning models and plotting data. Specifically for the analysis of huge steel manufacturing datasets, CeleraOne has created easy-to-use plotting methods.
Efficient Big Data Storage
Each supplier of industrial equipment and automation brings its own incompatible database. Conventional wisdom calls for the construction of a central data warehouse: time plus money plus internal resources. Buidling a data warehouse does not reduce production costs. State-of-the-art Machine Learning does not run on data warehouses.
CeleraOne supports steel manufacturers in moving along a sustainable route of Big Data handling. Centralization of data is required. It will be implemented efficiently and with substantial effects on production costs within limited timeframes.