Rapid process development using data generated in laboratory is important in chemical, pharma, specialty and biotech industries. Furthermore, growth in product demand has pushed these industries to adopt continuous process manufacture for new and existing processes. Further, in production, these processes have to be monitored and controlled to ensure safety and quality standards approved by US FDA. Hence, reaction processes are monitored using sensors either in offline manner (GC, HPLC, GC-MS) or online manner (spectrometers, temperature, calorimetry). Normally, offline data are obtained with delay. Hence, these measurements give rise to multi-sensors and multi-scale data. The proposal deals with developing online data analysis methods for rapid model identification and model update. Unsupervised (Principal component analysis, non-negative matrix factorization etc. ) or supervised learning methods (principal component regression, support vector machines) etc. are often applied to analyze online data. However, these methods suffer from rotational and scaling ambiguity. Hence, these methods cannot be applied directly to reaction systems due to underlying physical processes. In this proposal, we will propose to develop methods which combines a priori knowledge available regarding the measurements) and online data for building predictive and interpretable models. Further, we develop a method for online optimal input design to collect data in order to maintain the identified model. The proposed methods will be demonstrated with simulation as well as experimental data.