State Detection System Using Frequency Spectrum Pseudo Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing state detection systems for infrastructure and plant facilities face challenges in accurately and reliably detecting facility states using image recognition, due to limitations in image resolution and the inability to utilize high-resolution sensor signals effectively, especially when dealing with varying phases and phase differences in sensor signals.
Innovation Solution
A state detection system that converts digitized time-series signals from sensors into data on frequency spectrum intensity, reduces the expression word length of this data to low bits, and generates a pseudo image, which is then classified using image recognition to detect the facility state, thereby overcoming the limitations of image resolution and phase variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If time-series sensor signals are converted into pseudo RGB images for image recognition analysis, then the detection ability is improved and facility problems can be detected, but phase variations and phase differences in sensor signals cause infinite pattern variations in the converted images, reducing learning accuracy and inference accuracy
Solution Approach 1:
The patent converts time-series sensor signals from the time domain to the frequency domain using Fast Fourier Transform (FFT). This parameter transformation eliminates the influence of phase variations and phase differences, converting infinite pattern variations into a fixed frequency spectrum representation that maintains consistent learning and inference accuracy while preserving detection ability.
2Measurement precision
If high resolution sensor signals are used to detect facility states, then detection accuracy is improved, but incorporating image recognition AI into edge devices requires reduced expression word length, preventing full utilization of high resolution sensor signals
Solution Approach 1:
The patent extracts only the necessary frequency spectrum intensity information from the high-resolution time-series sensor signals and converts it to low-bit representation. By taking out the essential features (frequency components) and discarding redundant information (phase information and excessive precision), the system achieves accurate facility state detection while reducing expression word length for edge device compatibility.
3Reliability
If dedicated analysis algorithms are manually devised for time-series sensor signals, then detection reliability is improved, but system building time increases and cost increases
Solution Approach 1:
The patent copies the successful approach of converting signals to a different domain (time domain to frequency domain) and representing them as images, allowing the use of pre-trained image recognition AI models. This copying of the representation method enables reliable facility state detection without requiring manual development of new algorithms for each sensor type, significantly reducing system building time and cost.
Data Source
AI summary
A state detection system includes a computation unit that acquires data detected by a sensor. The computation unit includes a processing device that executes data processing. The processing device converts data of a digitized time-series signal from a sensor, into data on frequency spectrum intensity. The processing device converts a partial area in an overall area in which values of data on frequency spectrum intensity are distributed, into data expressed in low bits. The processing device generates a pseudo image, based on the data expressed in low bits. The processing device classifies the pseudo image, based on image recognition, and outputs a result of classification of a state of a facility.