PwC helps chemical company reduce unplanned manufacturing downtime by 25%

Unplanned production downtimes – often resulting from blockages, clogging or breakdowns of the production line - present a significant challenge for the manufacturing industry as they disrupt operations, increase maintenance costs and affect product quality and production schedules, thereby impacting overall operational efficiency and profitability. This was the case for a chemical client that was specifically experiencing blockages. The resulting loss exceeded double-digit millions due to global pipe fouling.

We worked with the client using a structured Lean Six Sigma DMAIC (define, measure, analyse, improve and control) methodology to define the problem at global level, incorporating insights from multiple plants through interviews and surveys, before narrowing down to a specific local scope and use case. The use case chosen was the plant experiencing the highest level of losses. The aim was to effectively reduce unplanned downtimes by integrating operational excellence with digital innovation.

The team used IoT sensor data and machine learning models to create early warnings and forecasting for cleaning and maintenance procedures. The production line had been operating with one or more blocked exhaust pipes for nearly half of the time, which was reduced to almost zero, leading to improved product quality and a 25% reduction in unplanned downtime.

All the client's plants are equipped with advanced distributed control systems (DCS), connecting them and enabling data from all plants to be consolidated onto a single digital platform, hosted on an Amazon web services (AWS) platform. Although multiple digital tools were being used on this platform, there was still unexploited data. 

Underutilisation of large amounts of data is common in companies; data is simply stored in the cloud or a system and not subjected to thorough data science analysis, meaning no KPIs are effectively derived from it and it’s not leveraged for deeper insights. Unlocking the full potential of this data requires a pragmatic approach to move towards informed decision making and drive significant operational improvements.

Developing a reliable prediction model was a multi-step process requiring attention to detail and a structured approach. We began with a measurement system analysis (MSA) to ensure the accuracy and reliability of the data being used, a foundational step that was crucial for establishing confidence in subsequent analyses. Following the MSA, extensive data collection and cleaning were conducted to eliminate noise and anomalies securing a comprehensive and pristine data set.

Correlation analysis identified trend relationships between various influence and production parameters and the occurrence of clogging events, as measured by pressure metrics. This was instrumental in narrowing down the variables that significantly impact the issue. Additionally, principal component analysis (PCA) was used to simplify the data by transforming it into a set of orthogonal components. This advanced technique helped further refine the variables and reduce the size of the data set.

Various predictive modelling approaches were explored, including time series forecasting, random forest and boosted regression models. The boosted regression model provided the capability to capture complex, non-linear relationships within the data. Time series analysis used historical data trends to predict future clogging events and proved to be the most suitable model for this application. The chosen long short-term memory (LSTM) time series model effectively captured long-term dependencies due to its recurrent neural network design. It also provided valuable temporal insights, making it essential for predicting and addressing clogging issues before they escalate.

The prediction model was seamlessly implemented and integrated into the plant's digital ecosystem, with Trendminer software used to visualise live predictions. This integration enabled real-time monitoring and prediction capabilities, empowering visual management and allowing plant personnel to proactively identify and address potential clogging issues before they impact operations. The use of a proof of concept (POC) approach was guided by a clear business case. This strategy proved to be both cost-effective and easily scalable, ensuring smooth deployment across the facility. 

PwC’s structured approach with clear process understanding, data capabilities and calling on cross-functional teams were the critical ingredients for the successful reduction of unplanned downtimes by 25% for the client. In addition, product quality improved due to less contamination of clogged products. Data availability, in terms of determining and implementing additional data captures, as well as connecting existing data sources, also proved key. Cross-functional staffing was also critical to tackle the diverse set of capabilities and engage different stakeholders. 

The combination of a solid methodology and predictive data models offers great potential for the further scalability of the clogging solution across manufacturing plants. More importantly, it perfectly addresses other issues affecting industrial performance and profitability, such as deviating environmental emissions and quality out-of-specs, energy consumption variability due to heat exchanger fouling, pump breakdowns, valve malfunctioning or clogged filters.