PSA Oscillation Detection via Discrete Fourier Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to accurately identify the root cause of faults in cyclical asynchronous production processes, particularly in multi-step asynchronous production units like PSA plants, due to difficulties in processing steady-state production variables and distinguishing between self-sustained oscillations and noise.
Innovation Solution
A monitoring and analysis system that uses a sensor array to collect data from production unit components, applies a discrete Fourier transform to generate frequency domain data, and compares amplitude ratios across units to detect abnormalities, defining operational limits and generating alarms for faulty components, which can be communicated to a control system for automated action.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If discrete Fourier transform is applied to detect oscillations in cyclic production units, then oscillation detection capability is improved, but difficulty in identifying root cause of faults increases due to phase coupling and multiple interacting loops
Solution Approach 1:
The patent segments the complex plant-wide oscillation problem into individual production unit analyses. By examining each cyclic production unit separately and comparing its oscillation characteristics against a library of known fault patterns, the system identifies root causes at the unit level rather than attempting to analyze the entire interconnected system simultaneously. This segmentation approach converts an intractable complex system problem into manageable individual unit diagnostics.
Solution Approach 2:
The patent transforms the analysis from time-domain signals to frequency-domain characteristics using discrete Fourier transform. By changing the parameter domain from temporal to spectral, the system can identify characteristic oscillation frequencies and patterns that indicate specific fault types. This parameter transformation enables the system to distinguish between different fault conditions based on their unique frequency signatures.
2Loss of time
If monitoring system continuously analyzes all production units, then fault detection speed is improved, but computational complexity and system resource requirements increase
Solution Approach 1:
The patent pre-computes and stores a library of expected oscillation patterns for various fault conditions and normal operating states. By having these reference patterns prepared in advance, the system can quickly compare real-time measurements against known patterns without performing complex real-time analysis. This preliminary preparation significantly reduces the computational burden during actual fault detection while maintaining high detection speed.
Solution Approach 2:
The patent uses simplified oscillation metrics and comparison algorithms that require minimal computational resources. Rather than implementing complex machine learning models or advanced signal processing techniques, the system employs straightforward frequency domain analysis and pattern matching that can be executed quickly with modest computational power. This approach prioritizes speed and simplicity over exhaustive analysis.
3Measurement precision
If system distinguishes between self-sustained oscillation and noise, then measurement precision is improved, but difficulty in detecting and measuring increases due to overlapping characteristics
Solution Approach 1:
The patent incorporates feedback from multiple measurement points and time intervals to characterize oscillation patterns. By continuously monitoring and comparing oscillation characteristics across different production units and time periods, the system builds a more robust understanding of what constitutes genuine oscillation versus noise. The feedback mechanism allows the system to adapt its detection thresholds and criteria based on observed patterns, improving discrimination accuracy over time.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This system enables early detection of minor problems, reduces diagnosis time for plant trips, and maintains optimal PSA performance by identifying statistically significant deviations in a multi-step cyclical asynchronous production process, thereby increasing productivity and reliability.
Implementation Method 1
A discrete Fourier transform is applied to generate frequency domain data
Data Source
Figure 1
Figure 1A
Figure 2
AI summary
Apparatus and methods are disclsosed that allow for the monitoring and analysis of production process data for a multi-step asynchronous cyclic production process (e.g. pressure swing adsorption) in a steady state plant (such as a steam methane reforming plant). Data collected from cooperating sensors is processed applying a moving window discrete Fourier transform (DFT). The transformed data can be further analyzed in the broader steady-state plant environment to accurately detect any process anomalies and avoid false alarms.