Industrial Internet State Detection With Adaptive Neural Depth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing industrial internet systems face low state detection efficiency due to high model complexity and increased computing power requirements, which affects production efficiency and poses security costs, especially under conditions of insufficient computing power at the edge.
Innovation Solution
A system state detection method using a shallow-layer feature extraction module followed by sequential deep-layer feature detection modules, where detection is terminated at a shallower level if satisfactory, reducing full-depth module usage and calculation amount, thereby optimizing detection efficiency under limited computing power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning models with increased depth are used for system state detection, then detection accuracy is improved, but detection efficiency deteriorates
Solution Approach 1:
The patent segments the deep learning model into multiple detection modules with different detection depths. Each module can independently perform detection at its own depth level, allowing the system to achieve high accuracy when needed while maintaining efficiency by using shallower detection for routine monitoring. This segmentation resolves the contradiction by providing multiple detection granularities rather than relying on a single deep model for all cases.
Solution Approach 2:
The patent implements dynamic adjustment of detection depth based on system conditions. The detection modules can adaptively select their operation depth according to the complexity of the current system state, switching between shallow and deep detection levels. This dynamic approach allows the system to maintain high detection efficiency for normal states while achieving high accuracy when anomalies are detected, thus resolving the efficiency-accuracy tradeoff.
2Measurement precision
If full-depth deep learning models are used for all detection cases, then detection accuracy is improved, but computing power consumption increases
Solution Approach 1:
The patent applies partial action by using different detection depths selectively rather than always employing full-depth models. For routine monitoring and simple cases, shallower detection modules are used, consuming less computing power. Full-depth models are activated only when necessary for complex or ambiguous cases, achieving high accuracy without the continuous high energy consumption of always using deep models.
Solution Approach 2:
The patent changes the detection depth parameter dynamically based on system conditions and data characteristics. By adjusting this parameter, the system can optimize the balance between accuracy and computing power consumption for different operational scenarios, using minimal computing resources while maintaining necessary detection accuracy.
3Productivity
If shallow-layer detection is used, then detection efficiency is improved, but detection accuracy deteriorates
Solution Approach 1:
The patent segments the detection system into multiple modules with progressively deeper detection capabilities. This segmentation allows the system to use shallow detection for efficiency-critical scenarios while having deeper modules available for accuracy-critical scenarios, resolving the contradiction by providing a spectrum of detection depths rather than a single fixed level.
Solution Approach 2:
The patent implements dynamic selection of detection depth based on the specific detection task and system state. When efficiency is the priority (e.g., real-time monitoring of stable systems), shallow detection is used. When accuracy is prioritized (e.g., diagnosing complex failures), deeper detection is activated. This dynamic adaptation resolves the static contradiction between efficiency and accuracy.
Data Source
AI summary
A system state detection method and system for an industrial internet system, a device and a medium, all related to the field of industrial internet. The method includes: acquiring system data of the industrial internet system; detecting the system data by using a system state detection model to obtain a system state of the industrial internet system; where the system state detection model includes a shallow-layer feature extraction module and a plurality of deep-layer feature detection modules connected in sequence, the system state detection model is obtained by training a dynamic neural network model by using a training data set, the training data set includes historical system data of the industrial internet system, and the system state includes system normality and system abnormality; and in a case that the system state is system abnormality, outputting the system state of the industrial internet system.