Industrial Plant Action Sequencing for Abnormal Operating States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial plants face challenges in automatically remedying abnormal operating states, as existing systems like DCS can only output alarms and require operators to execute specific sequences of actions, which can be complex and error-prone, especially in safety-critical situations with limited training data.
Innovation Solution
A computer-implemented method using state encoder and action decoder networks to encode operational state variables into a lower-dimensional representation, mapping this to a sequence of actions, which can be automatically executed or notified to operators, aiding in remedying abnormal situations by highlighting necessary control elements and ensuring safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the DCS outputs alarms for abnormal conditions, then the operator is notified of the abnormal state, but the operator still must manually execute complex sequences of actions which is error-prone and time-consuming
Solution Approach 1:
The patent introduces an intermediary system (the sequence determination system) between the alarm output and the operator. This system automatically determines and provides the correct sequence of actions to be taken, serving as a mediator that guides the operator through the remediation process rather than relying on the operator's memory or manual lookup of procedures.
Solution Approach 2:
The system enables self-service by automatically generating and providing the action sequence based on the detected abnormal condition. Instead of requiring the operator to manually determine the correct actions, the system serves itself by computing the appropriate response sequence and presenting it to the operator for execution.
2Measurement precision
If the system uses a full state-to-action network to map all state variables directly to actions, then complete information is available, but training becomes computationally intensive and data-intensive
Solution Approach 1:
The patent segments the state-to-action mapping process into two distinct stages: first encoding state variables into a compressed representation, then mapping this representation to actions. This segmentation allows the system to handle high-dimensional state variables without requiring a single complex network to process all variables directly to actions, thereby reducing training complexity while preserving information.
Solution Approach 2:
The patent introduces an intermediate dimensional space (the compressed representation) between the high-dimensional state variables and the action space. By transforming the problem from a direct high-dimensional to low-dimensional mapping to a two-stage process involving a compressed intermediate representation, the system reduces the computational burden of training while maintaining the ability to accurately characterize system states.
3Productivity
If the system compresses state variables into a lower-dimensional representation, then training becomes easier and faster, but information loss may occur
Solution Approach 1:
The patent performs preliminary encoding of state variables into a compressed representation before the state-to-action mapping step. This preliminary action of compression is designed to retain only the most relevant information needed for determining actions, effectively performing feature selection or extraction beforehand to prevent information loss in the subsequent mapping stage.
Solution Approach 2:
The system incorporates feedback mechanisms where the compressed representation is evaluated based on its ability to predict appropriate actions. The encoding process can be refined through feedback from the action prediction accuracy, ensuring that the compressed representation retains sufficient information for effective decision-making while maintaining computational efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method (100) for determining an appropriate sequence (6) of actions to take during operation of an industrial plant (1) or part thereof, comprising the steps of: • obtaining (110) values of a plurality of state variables (2) that characterize an operational state of the plant (1) or part thereof; • encoding (120), by at least one trained state encoder network (3), the plurality of state variables (2) into a representation (2a) of the operating state of the plant (1) or part thereof; • mapping (130), by a trained state-to-action network (4), the representation (2a) of the operating state to a representation (6a) of a sequence (6) of actions to take in response to the operating state; and • decoding (140), by a trained action decoder network (5), the representation (6a) of the sequence (6) of actions to the sought sequence (6) of actions to take.