Dual Neural Network Ignition Timing Control for Knocking Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ignition timing control systems for internal combustion engines face challenges in accurately detecting knocking intensity due to mechanical vibrations, leading to erroneous delays in ignition timing, which can result in reduced engine output and efficiency, especially in commercial vehicles with varying engine components and vibrations.
Innovation Solution
A dual neural network system is employed, where a first learned neural network estimates knocking intensity from pressure sensor data and a second neural network predicts the decrease in this estimate when ignition timing is delayed, allowing for precise control of ignition timing adjustments based on the difference between the two networks' outputs to prevent excessive delay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a knocking sensor is used to detect engine body vibration, then knocking intensity can be detected, but mechanical vibrations from valve seating and fuel injection cause erroneous determination of high knocking intensity
Solution Approach 1:
A neural network is introduced as an intermediary between the knocking sensor output and the ignition timing control. The neural network processes the sensor signals to distinguish between mechanical vibrations and actual knocking events, filtering out false positives while preserving true knocking detection capability.
Solution Approach 2:
The system creates a learned model (copy) of the relationship between knocking sensor signals and actual knocking intensity through training data. This learned neural network copy can replicate accurate knocking detection without being directly affected by mechanical vibrations that confuse the raw sensor output.
2Measurement precision
If a pressure sensor is used to detect combustion pressure for accurate knocking intensity detection, then knocking intensity can be accurately detected, but the pressure sensor is expensive and deposits adhere to it causing combustion form deformation
Solution Approach 1:
The system replaces the expensive, maintenance-prone pressure sensor with a cheaper knocking sensor combined with a computationally processed neural network model. This substitution uses a less costly component that doesn't suffer from deposit adhesion and combustion form deformation issues.
Solution Approach 2:
The physical pressure sensing mechanism is replaced with a signal processing system using neural networks. Instead of mechanically measuring combustion pressure directly, the system uses learned patterns from knocking sensor data to infer knocking intensity, substituting a mechanical measurement system with an information processing system.
3Ease of manufacture
If a learned neural network is used to estimate knocking intensity from knocking sensor output, then cost is reduced, but the neural network cannot learn engine vibration specific to each commercial vehicle causing incorrect determination
Solution Approach 1:
The neural network is trained in advance using teaching data that includes various engine vibration patterns and knocking characteristics. This preliminary learning phase allows the network to acquire knowledge about different vibration sources before actual operation, enabling it to distinguish between mechanical vibrations and knocking without needing to adapt during runtime.
Solution Approach 2:
The system uses feedback from the knocking sensor and pressure sensor (during training phase) to continuously refine the neural network's understanding of engine-specific vibration patterns. By incorporating actual engine operation data into the training process, the network learns to adapt to individual engine characteristics while maintaining cost-effectiveness.
Data Source
AI summary
An ignition timing control device for an internal combustion engine includes a storage device and a processor. The storage device stores a first learned neural network and a second learned neural network. The processor is configured to perform, in a next cycle where ignition timing is delayed, control to delay the ignition timing in a cycle after the next cycle based on a difference between a predictive value of an estimate of a value representing knocking intensity calculated with use of the second learned neural network and the estimate of the value representing the knocking intensity calculated with use of the first learned neural network. When the difference is larger than a predetermined set value, the processor is configured not to perform the control to delay the ignition timing in the cycle after the next cycle.