Partial Neural Network Control for Real-Time Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in manufacturing systems, particularly in complex systems that combine process and discrete manufacturing, face inaccuracies due to their design for specific types of manufacturing, and current neural network approaches are computationally expensive and impractical for real-time anomaly detection.
Innovation Solution
A partially connected neural network topology is developed based on event ordering relationships, where connections between neurons represent the frequency of subsequent occurrences of events, reducing network complexity and improving accuracy by mimicking the connectivity within the manufacturing system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fully connected neural network is used for anomaly detection in complex manufacturing systems, then the detection accuracy may be improved, but the computational complexity and training time increase significantly
Solution Approach 1:
The patent segments the fully connected neural network into multiple layers with specific connectivity patterns. The input layer receives signals from multiple sources representing different manufacturing events, and these are processed through hidden layers with defined connection structures before producing anomaly detection outputs. This segmentation reduces the computational burden while maintaining detection capability.
Solution Approach 2:
The patent applies local quality by creating a partially connected neural network where not all neurons are connected to all neurons in adjacent layers. Instead, connections are established based on the event ordering relationships specific to each manufacturing system. This localized connectivity structure reduces the total number of parameters and computational operations while preserving the ability to detect anomalies relevant to the specific manufacturing context.
2Measurement precision
If a fully connected neural network is used for anomaly detection, then comprehensive pattern recognition is achieved, but the training time becomes excessively long for real-time applications
Solution Approach 1:
The patent extracts and removes unnecessary connections from the fully connected neural network structure. By analyzing event ordering relationships in the manufacturing system, the patent identifies which connections are essential and which can be removed. This extraction process reduces the training time by decreasing the number of parameters to optimize while retaining the critical patterns needed for accurate anomaly detection.
3Ease of manufacture
If anomaly detection methods designed for specific manufacturing types are applied to complex systems combining process and discrete manufacturing, then the method is simple to implement, but the detection accuracy decreases
Solution Approach 1:
The patent creates a universal anomaly detection framework that can handle both process and discrete manufacturing within a single neural network architecture. The system accepts multiple types of input signals representing different manufacturing events and uses a standardized partially connected neural network structure that adapts to the specific event ordering relationships of any manufacturing system type. This universal approach maintains implementation simplicity while improving accuracy for complex hybrid manufacturing systems.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An apparatus for controlling a system including a plurality of sources of signals causing a plurality of events includes an input interface to receive signals from the sources of signals, a memory to store a neural network trained to diagnose a control state of the system, a processor to submit the signals into the neural network to produce the control state of the system, and a controller to execute a control action selected according to the control state of the system. The neural network includes a sequence of layers, each layer includes a set of nodes, each node of at least an input layer and a first hidden layer following the input layer corresponds to a source of signal in the system. A pair of nodes from neighboring layers corresponding to a pair of different sources of signals are connected in the neural network only when a probability of subsequent occurrence of the events in the pair of the different sources of signals is above a threshold, such that the neural network is a partially connected neural network.