Beyond Traditional Chemometrics: The AI Advantage

Advances in process control technology enabled through the recent progress in AI and Machine Learning allow further improvements in operating efficiency through more capable APC (Advanced Process Control) solutions. The need for precise process control is driven by the pressure for transition to sustainable production, increasingly stringent emissions control regulations, ever more complex process technologies, and tight profit margins in raw materials price volatility conditions. The APC market growth shows the importance of process control in today’s industry. Based on the report by Reports Insights (2023), the advanced process control (APC) market is estimated to reach over USD 3,407.95 million by 2030, compared to USD 1,673.29 million in 2022. The market is projected to grow at a Compound Annual Growth Rate (CAGR) of 9.6% from 2023 to 2030.

Online analysis, specifically NIR spectrometry, plays a significant role in maximizing the potential of APC in improving the technological process efficiency. Measuring the feedstock properties allows timely adaptation of the process parameters to the ongoing fluctuations in the raw material compositions. Direct measurement of the process product properties using NIR analyzers can be a more reliable alternative to soft-sensors within the APC sensor. This added value of NIR analyzers to process control and process health measurement explains the trend for growing global NIR spectrometry market. According to Grand View Research (2022), this market was valued at USD 502.7 million in 2022 and is expected to expand at a compound annual growth rate (CAGR) of 5.8% from 2023 to 2030.

These benefits of NIR analyzers to process control can only be unlocked if adequate and updated spectrometric models are deployed on them. The problem with the current spectrometric modelling tools is two-fold. Firstly, they require high level of user proficiency to train the models based on available lab results with their corresponding spectra. Secondly, they don’t support live-update functionality, allowing automatic model update, once more reference data becomes available.

Process NIR analyzers are known for their simplicity of installation, high reliability, and minimal maintenance requirements. However, it is important to note that this technique is not linear and may not provide clear resolution for all components in the sample. Consequently, any changes in stream composition or process conditions necessitate model updating. Therefore, reliability of process NIR analysis heavily relies on the accuracy of chemometric models.

New challenges in the processing industry necessitate precise and adaptive Advanced Process Control (APC). Adequately controlling the process is only possible when its present state is correctly determined. Online analysis in the form of NIR spectrometry plays a crucial role in allowing this. The bottleneck in exploiting the predictive power of NIR spectrometry is keeping the its models up-to-date as it requires skillset not always available in the processing industry. The new MODCON.AI auto-modelling software uses the methods of AI to make spectrometric modelling accessible to wide circle of practitioners without specialized mathematical background.

Chemometrics is a fascinating technology allowing real-time non-destructive analysis of the material to identify its physical and chemical properties of the material. This technology is an alternative to laboratory analysis, which while valuable and necessary, cannot support real-time decision-making for steering the process and monitoring its health. NIR spectrometry relies on chemometric algorithms for instant material properties identification in petrochemistry, biotechnology, pharmaceutics, and food production industries.

MODCON.AI auto-modelling software addresses these limitations of the existing spectrometric modelling solutions. In building the initial model from the array of the available reference data, MODCON.AI harnesses the recent achievements of AI technology to replace or facilitate the user’s judgement. The automated modelling performance approaches and even exceeds the accuracy of the models created by an expert practitioner. The auto-modelling software also allows live update, incorporating the new lab results into the model as soon as they become available.

Application Use Cases:

1. Initializing a model for the new NIR analyzer deployment using the full set of available calibration data.
2. Updating the existing NIR analyzer predictive model making use of the increasing over time pool of available calibration data.

Below, you'll find an illustration of the performance report that highlights the advantages of this methodology for automatically building and updating chemometric models.

The introduction of MODCON.AI auto-modelling software addresses a significant bottleneck in utilizing the value NIR analyzers provide to the processing industry. Outdated models lead to unreliable spectrometric predictions that can’t be used for APC. Therefore, the impact of deployment of high-quality models on the NIR analyzers is underpinning process efficiency, sustainability and the quality of its products.