Emerging Trends in Automation – AI & ML
Published on : Friday 04-12-2020
Artificial Intelligence and Machine Learning are going to be major contributors to the technological disruption in industrial automation, says Nitin Sambhare.
As per global studies, unexpected disruptions in industrial space can cost 5-8% of capacities. It leads to an annual cost of US$20+bn in the US alone in terms of lost production, equipment damages, environmental impacts, human casualties, litigation cost, insurance claims, etc. Industrial automation is going to witness a significant technological disruption over the next decade with a major contribution from the Artificial Intelligence and Machine Learning technologies to optimise handling of industrial processes with minimal losses.
Traditionally, industrial processes are automated using PLC/DCS across continuous/batch processes and discrete manufacturing spaces to optimise production and minimise production disruptions. Basic level automation is achieved through simple and medium complex control strategies in Continuous Process, recipes in Batch Process and sequential logic controls in Discrete Manufacturing.
With further technology advancement, Advanced Process Control (APC) got introduced with high level computing controllers and mathematical modelling using multivariable control strategies. Prior to this, it was more of linear controls using 2-4 process variables. So far, APC was thought of as the best available automation possibility in Continuous and Batch processes for achieving optimum productivity/yield within known constraints. While it achieved this objective with a high level of success, it needed a huge amount of effort in terms of building mathematical models using process dynamics and fine-tuning the model using step-tests on the running plants. Moreover, the model thus developed was mostly limited to process dynamics and equipment behaviours at that moment and the accuracy of the model deteriorated as the equipment performance changed over a period of time due to wear and tear, and thus the process dynamics. Furthermore, due to the huge efforts required to remodel/recalibrate, it was not possible to rebuild the model each time during run time.
With advent of AI/ML technologies, it is now possible to overcome most of the lacunas with APC systems, and over and above, it can practically replicate the collective human intelligence in the most efficient way. It is opening up many more possibilities in the industrial automation field, which nobody would have thought of a few years ago.
Key differentiators – APC vs AI/ML based solutions
a. Fixed model based on data collected Vs Self learning model updated on real-time data
b. Platform dependent Vs open source algorithms independent of platforms
c. Dependency on suppliers for future updates Vs self-learning mechanisms
d. High cost of owning hardware/software dependent solution Vs reasonable cost of owning Open Source platform, interoperability and in-house skill development
e. Rigid in terms of flexibility to adopt to the future changes Vs flexible to align with day to day changes, and
f. Mostly works within known constraints Vs works across known and unknown constraints.
Challenges in Industrial space
1. To maximise expected outcome from the industrial process in terms of Yield/Throughput, etc
2. To minimise abnormal situations which can lead to uncontrolled processes resulting into risk to human/environment/community/property, etc., and
3. To optimise performance of industrial equipment in terms of availability, overall equipment efficiency, energy consumption, etc.
What is AI/ML based solution?
AI/ML based solution basically starts with developing algorithms using real-time data that the industrial process exhibits during the run time and building a mathematical model that represents the process dynamics in a most accurate way. In its raw form, the model might not be useful for optimising the process performance as it will have many deviations from real-time performance/behaviours. However, by applying machine learning principles, the model can be taught to learn the exact behaviour of the industrial process in the real-time.
Some of the common methodologies used during the process of building AI/ML based solution:
i. Linear Regression: is a linear approach for modelling the relationship between a controlled process variable and one or multiple manipulated process variables (Refer Fig.1).
ii. Logistic Regression: is a statistical model, in its basic form, uses a logistic function to model a binary controlled process variable, estimating the parameters of a logistic model. It is a classification algorithm used to predict a binary outcome given a set of manipulated variables. This algorithm can be used to detect anomalies in the process or equipment (Refer Fig.2).
iii. Support Vector Machine: These are supervised learning models with associated learning algorithms that analyse data used for classification and regression analysis. It tries to finds the ‘best’ margin (distance between the line and the support vectors) that separates the classes and reduces the risk of error on the data.
iv. Time series algorithms: Time series data can be phrased as supervised learning. Given a sequence of numbers for a time series dataset, we can restructure the data to look like a supervised learning problem. We can do this by using previous time steps as input variables and use the next time step as the output variable.
v. Random Forest: Random forests are an ensemble learning method for classification, regression and other tasks that operate by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes or mean/average prediction of the individual trees.
Some of the key industrial applications using AI/ML based solutions
A. Process unit operation – All the process plants consist of different equipment (static and rotating) such as pumps, compressors, etc. These are operated by the field personnel to run the process smoothly and are required to be changed over to stand-by ones for maintenance/break-down related issues, specifically while handling the rotating equipment, while process is expected to run without disturbance. These tasks become critical when a small change in the process parameters (pressure/flow/temperature, etc), can impact the overall process significantly. In such cases, the field operator’s skill/ability in carrying out the tasks smoothly is very crucial. Generally these skills/experience vary from person to person, hence the process disturbance while carrying out these tasks also vary significantly, leading to unpredictable/unplanned production losses.
AI/ML based solutions using Time Series and Logistic Regression algorithms along with Random Forest Classifiers can be used to automate entire sequence of tasks that the field operator is supposed to execute in a full or semi auto mode, achieving the best possible sequence of task execution in these kinds of mission critical tasks. For example, if the running compressor delivering H2 to a process at 100+ psig pressure needs to be changed over to stand-by compressor, any small fluctuations in the load/pressure during the changeover can lead to production disturbances including tripping of the plant. Using AI/ML, the algorithm can be built based on real-time data and past events from such changeovers by the experienced field operators. The algorithm can then put in use for identifying the best possible sequence to be followed during such tasks. The algorithm will then continuously keep learning the complexity & relationship amongst the various tasks, parameters, events, etc., and can be used to build a model which could be a replication of the best operator’s experience in handling those tasks. If the same model can be put in the operation, even a less experienced operator can repeatedly produce the same result as the most experienced operator, leading to less production losses/process disturbances.
B. Golden Batch – Another example of AI/ML based solution is to use Time series, Clustering and Regression algorithms including Support Vector Machines together to identify ‘Golden Batch’ in the batch process recipes where the process is able to run at its best possible efficiency and then build a model using the data captured in the Golden Batch. The model built in such a way can then execute all recipes to achieve the same level of quality and productivity across all batches irrespective of personnel involved in the operation. The model further can be fine-tuned during run-time as it can keep learning itself using machine learning principles.
C. Prediction model using Support Vector Machines/Random Forest – These methodologies can be put to best use to predict plant operation key performance indicators in a complex industrial process in a real-time basis. Generally process lags and dead time along with complex equipment performance behaviours make it difficult to predict the performance in a timely manner. For sure, the operation team is able to do near perfect analysis of the events and causes after the events have happened. But it doesn’t help much to avoid such events in a realistic manner. AI/ML based solution can effectively predict these undesirable events in a timely manner and guide the operation personnel to take appropriate actions to avoid it. In fact, most advanced solutions using the AI/ML principles can intervene in the process to minimise the impact even if the operation personnel don’t respond to such predictions in a timely manner.
With such newer approaches, there is a huge possibility of finding better solutions to the industrial automation problems which remained unresolved so far.
Nitin Sambhare is Managing Director, Desmet Ballestra Engineering Centre Pvt Ltd, located at Bangalore. His work experience spans over 30 years, mostly devoted to engineering and design fields across industrial domains in Continuous/Batch process and Discrete manufacturing space. He has been instrumental in building global engineering centres across India and China for delivering world class engineering and design solutions to global customers. Prior to his present association, he had been associated with companies like Reliance Industries Limited, Honeywell, and Rockwell.