Measuring process improvements with Advanced Process Control
Advanced Process Control (or APC) adds an extra control layer with the aim to improve plant production and energy performance.
APC is an additional control layer on top of basic single input and single output control. APC can have multiple inputs and outputs that it evaluates together and combines this with the capability to do short term predictions in real-time. Adding the APC layer requires great expertise and adds complexity to your process and so, what are the real-world benefits?
What problem does APC solve?
A simplified process overview is shown below where the plant receives raw product from stockpiles for processing and it is measured by weightometers on the feed conveyor. The raw material is then processed through a couple of stages and the resultant streams are the saleable product (about 60% of the raw feed) and waste. Energy consumption is metered by energy meters on the main supply to the plant.
The plant operators need to monitor and control several circuits continuously and are in constant contact with field personnel doing checks and maintenance. With many manual controls that needs to be managed, operators simply cannot operate the plant optimally all the time and plant performance can also vary depending on operator experience.
The aim of APC is to stabilise the feed into to plant and then drive production towards the plant’s design capacity, by automating the manual controls.
Real-world data is messy
Historic process data was initially available only in daily intervals, recorded in spreadsheets by the operators using a daily totaliser tag from the weightometers. As part of the project, a data historian was implemented and thus process data was now historized at one-minute intervals. This was done together with the implementation of APC and thus limited granular historic data is available. Energy data is available from the energy metering system in half-hour intervals with long-term history.
For the analysis of the process, the more granular data was used from the historian as more detailed process data was needed. For the energy performance analysis, rolled up daily data was used to enable use of the manually recorded historic data. As the model will be based on the whole facility, it should allow for better models to be created as detailed process variables were not available to incorporate at this time.
Did the plant operators adopt APC?
The challenge with an APC implementation is the change of mindset in that one is not chasing the instantaneous metric of highest throughput (tons per hour), as seen in the control room, but more consistent throughput which will deliver, overall, more product over the day. APC was commissioned together with the operators in the control room to get maximum buy-in and provide on-the-job training.
APC operates within a range that is set by the operators. Sometimes operators fell back into old habits of trying to get more throughput as the APC was hitting constraints and they clamped the operating range. Meaning that the range did not provide sufficient space for the algorithms to stabilise and optimise, based on the current operating conditions.
Using the data from the historian, we can track the APC utilisation when the plant was running. APC utilisation is a key indicator to show the level of adoption by the operators as they need to manually switch on APC control after the plant has been started up.
During October 2019, APC was being commissioned and one would expect utilisation to be low during this period. After this period, average utilisation was above 50%, except for April 2020 and June 2020, the first being related to COVID lockdown and the second due to a plant shutdown.
Average APC utilisation indicates good adoption by the plant operators.
What were the benefits relating to production?
The data was filtered to include data when the plant was running (based on the process run signal) and was split into an APC off period and an APC on period (based on the controller mode). The comparison was done on the average of the two main feed weightometers into the plant.
There are periods where the process was running, but there was no product feed, and this is biasing the results more in favour of APC on. It is worth to note that the APC controller will switch on when the process has reached certain conditions. To correct for this, additional filtering was applied on the weightometers to only include data where the feed was more than 230 tons per hour (based on the lower whisker of the APC off data), which yields a fairer comparison.
From the plots and statistics, with APC control utilised, the process was more stable resulting in a reduction in standard deviation of 24.0% and an increase in throughput of 7.5%, from 739.2 tons per hour to 794.6 tons per hour.
There is only 12.2% APC off data which may bias the results, but smaller data selections shown comparable results. Another observation is that the APC off histogram shows two distinct modes as this is manual operation, while APC on shows a more left skewed distribution as expected. This is because the APC objective is to maximise throughput to reach the upper limit.
Process stability has improved, and plant throughput increased by 7.5%.
Does APC improve energy performance?
Good understanding of the process is essential to build a good model for energy analysis. This is a simple plant process with no heating or cooling involved thus external temperature is a negligible impact. Most of the energy consumption is influenced by the raw feed. Following Measurement and Verification protocols used in energy management, the baseline period was selected as the first nine months of 2019 as the first controller was implemented in early in October 2019.
Data from several sources needed to be combined to perform the analysis and there where quite a few obvious outliers that didn’t fit the process context. The analysis only considered the productive period, i.e. where production was more than 230 tons per day; however, there may be other energy saving opportunities in the non-productive period, but this is outside the scope of this analysis.
A regression model was developed that would predict the expected energy consumption from October 2019 onwards. The CUSUM is the cumulative sum of difference between the expected consumption and the actual energy consumption. A negative CUSUM trend (downwards) indicates that the plant is using less energy than expected, thus a saving. A positive trend indicate more energy is being consumed than expected while a horizontal line indicates than the expected consumption and actual energy consumption are the same (or at least similar).
From the CUSUM graph, a clear downward trend is observed indicating an improvement in energy performance of about 3% from when APC was implemented. However, from around May 2020, energy consumption increased significantly and was consistently more than expected (about 6%), indicated by the upward sloping trend.
The cause of this increase was related to additional equipment being added that was used for other processes not related to the plant; however, it was fed from the same plant electricity supply. In this case, a new baseline model needs to be created with the additional equipment taken into consideration by including the relevant variables or separately metering the additional equipment.
Energy performance improved by 3% during the first six months of operation.
Conclusion
With the implementation of APC, the plant ran more consistent and increased throughput. Operators have adopted the new control technology, giving them more time to focus on other important tasks. Better process efficiency resulted in improved energy performance, but the initial model is no longer representative from mid-2020 and need to be revisited.
The analysis is available in my GitHub repo.
