Common current approaches for analysing performance and predicting the piping stress analysis output require stress engineers to manually assimilate the traditional software output from experience and then spend hours, if not days, in bringing the system within the limits of the design criteria. While this approach works on less complex piping systems, too much data wrangling and cognitive analysis could result in not only overlooking the optimal solution, but also lead to redundant iterations. Lack of practice, abilities, experience, insight, and intuition in assessment of the software output, especially pattern recognitions and inability to think critically & alternatively – supress insights into the failure behaviour of the piping system and conceal signals that are pointing to an upcoming failure.
As pattern recognition is the core of all machine learning algorithms, is it possible to identify patterns using machine learning through a gathered data? Could collaboration of a human proficiency and historical structured datasets subsequently drive decision making to result an improved turnaround, ensuring safety of connected equipment and the supporting piping structure?
Addressing Issues of Traditional Software Tools
Piping stress analysis is performed taking in view the pipe arrangement, support location & span and the support type by the cause and effect of many moderating variables such as process parameters, nozzle load limits, material properties, FEA, flange leakages, seismic and other static & dynamic loads etc. Present software for the analysis is based on strength of material concept of BEAM theory and stress and strain properties for a material and many advanced courses teach professionals to feed data in software – which provides result depending on implicit knowledge of the stress engineer and the input-operation-output trial and error. As the piping route and piping supports have to be re-modelled in the stress analysis software after any change in the layout, an intelligent system would need to break the design–analysis loop of the traditional use of analysis software.
Can current software tools for different design stages of process plants (e.g., process analysis, plot plan development, detailed design 3D modelling, and mechanical stress analysis), use historical data from completed and existing projects to build and compute the required intelligence for the early stages of a new project? Stress analysis done in a more efficient way would require a machine-learning-based approach with learned capability of the software system to use heuristics, optimize dataset on benchmark variables and simulate patterns, but more importantly – prompt feasible solutions by a predictive analysis model to determine whether route remodelling is needed. Let’s take a simple example of a ‘Column Re-boiler system’ in a refinery or a petrochemical project to explain this further.
It is challenging for stress engineers to decide optimal design as they run various iterations in software before finalizing one design. Depending on the process criteria, piping layout, constraints, available area, support feasibility and equipment nozzle detail – an algorithm can verify all the possible routes and piping supports between two points (equipment nozzles) in 3D space by automatically matching with a database of piping routes with their flexibility analysis results and build a predictive model of the best option. Artificial Intelligence can provide tacit knowledge and prompt errors or highlight strain or sensitive zones wherein corrective action could be taken on deciding the most optimum orientation, a more feasible support arrangement and its location, thereby accomplishing an optimal analysis.
Let us delve deeper into how a ‘learnt machine’ can automate some inputs to the analysis process. While the choice of ‘node locations’ is up to the designer, but can a ‘trained’ system accurately represent this piping system by automatically suggesting the nodes required at each change of direction, at all in-line equipment, and at all restraints (supports and terminal points)? The answer is, yes! The machine could also be further educated to include additional nodes as per the output (internally computed by its own algorithm); for example, at the midpoint between supports so that the deflection between the supports is brought within the allowable limits. Thus, machine learning or Artificial Intelligence can help stress engineers to do the cognitive analysis much faster and forming clusters of variables or recognising the pattern affecting the analysis.
Future of AI in Piping Stress Analysis
Traditional stress analysis software developers can establish algorithm logic to form, validate and suggest possible design scenarios for piping and support design in process plants – in effect to automate the stress analysis activity to effect decisions and gradually bring redundancy in current trial and error practice. With today’s digital realities of Industry 4.0, modelling non-linear problems and predicting the allowable stress values for given input parameters based on Deep Neural Networks using big data is possible. Neural networks rely on human knowledge integration or training data i.e. applicable codes & allowable stress, civil- mechanical-electrical-instrumentation interference, orientation, loads, support type & span, methods of construction, engineering best practices etc. to learn and improve their accuracy of the target attribute over time. However, once these sets of learning algorithms are fine-tuned, they are powerful tools in artificial intelligence, allowing us to classify and cluster data for a given set of inputs at a very rapid pace.
Algorithms can validate clash checks, safety distance checks, parallel equipment arrangement check, scrutinize different safety aspects in the piping and instrument diagram (P&ID) or process flow diagram (PFD) of the piping system. Additionally, algorithms can take into account validation of the analysis, or conversely, discrepancies for un-buildability elements, patterns for repeated failures on account of vibrations, leakages, material failures etc. from construction knowledge and import it back to stress analysis knowledge base.
In essence, Artificial Intelligence can be used for automatic generation of data & knowledge bases and build a structural model of most variables affecting the piping stress analysis and the data correlations, which are normally imperceptible to humans, and the patterns of which are impossible for human brain to simulate through.
References:
http://www.ijfis.org/journal/view.html?uid=875&&vmd=Full
https://becominghuman.ai/what-is-training-data-its-types-and-why-it-is-important-f998424c3c9
https://asmedigitalcollection.asme.org/OMAE/proceedings-abstract/OMAE2017/57731/V07AT06A043/281647
