SDAE-DBN Surface Roughness Prediction via Acoustic Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional surface roughness measurement methods face limitations in accuracy and efficiency due to the need for contact measurement tip maintenance and sensitivity to surface dirt, while AI-based methods require extensive labeled data and manual feature extraction, hindering on-line prediction capabilities.
Innovation Solution
An on-line prediction method utilizing a stacked denoising autoencoder-deep belief network (SDAE-DBN) that collects and processes vibration and noise signals from machining processes, employing polynomial least square and five-point cubic smoothing methods to normalize and extract features, reducing manual intervention and data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If contact measurement is used to measure surface roughness, then measurement can be performed without cleaning the surface, but the measuring tip is easy to wear and scratch the surface, limiting its application in high-precision detection
Solution Approach 1:
The patent replaces the mechanical contact measurement system with an acoustic field-based measurement system. By using acoustic emission signals generated during machining and processing them through signal processing and neural network algorithms, the system achieves high-precision surface roughness measurement without physical contact, thereby eliminating tip wear and surface scratching while maintaining measurement accuracy.
2Measurement precision
If non-contact measurement is used to measure surface roughness, then measurement accuracy is improved, but the method is sensitive to surface dirt and requires cleaning before measurement, reducing measurement efficiency
Solution Approach 1:
The patent performs preliminary action by capturing acoustic emission signals during the machining process itself, before the workpiece leaves the machining center. This allows surface roughness to be predicted in real-time without requiring subsequent cleaning or separate measurement operations, thereby maintaining high measurement accuracy while significantly improving measurement efficiency.
Solution Approach 2:
The patent introduces acoustic emission signals as an intermediary that correlates with surface roughness but is not affected by surface dirt. The acoustic signals generated during machining serve as a mediator that allows indirect measurement of surface quality without direct contact with the workpiece surface, eliminating the need for cleaning operations.
3Reliability
If traditional neural network methods are used for surface roughness prediction, then the model can learn from data, but manual feature extraction is required which makes the data processing process cumbersome
Solution Approach 1:
The patent applies self-service by implementing automatic feature extraction within the neural network model. The network automatically learns and extracts relevant features from raw acoustic emission signals through its hidden layers, eliminating the need for manual feature engineering. This reduces data processing complexity while maintaining or improving prediction accuracy through the model's autonomous feature learning capability.
4Reliability
If traditional neural network methods are used for surface roughness prediction, then the model can be trained with sample data, but a large number of labeled data are required which limits wide use in industry
Solution Approach 1:
The patent applies partial action by using a reduced set of labeled data for training the neural network model. By leveraging the automatic feature extraction capability and the inherent information in acoustic emission signals, the model achieves satisfactory prediction accuracy with fewer labeled samples, thereby reducing data acquisition costs and enabling broader industrial application.
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 approach enables real-time, high-accuracy surface roughness prediction with reduced reliance on labeled data and manual feature extraction, improving prediction efficiency and accuracy, and minimizing human error.
Implementation Method 1
the tri-axis acceleration sensor is adsorbed on the rear bearing of the machine tool spindle through the magnetic seat to collect the vibration signals of the cutting process
Implementation Method 2
a microphone is placed in the left front of the processed part to collect the noise signals of the cutting process of the machine tool
Implementation Method 3
the polynomial least square method is used to eliminate the trend term of the dynamic signal
Implementation Method 4
the five point cubic smoothing method is used to smooth the signal
Data Source
AI summary
An on line prediction method of part surface roughness based on SDAE-DBN algorithm. The tri-axis acceleration sensor is adsorbed on the rear bearing of the machine tool spindle through the magnetic seat to collect the vibration signals of the cutting process, and a microphone is placed in the left front of the processed part to collect the noise signals of the cutting process of the machine tool; the trend term of dynamic signal is eliminated, and the signal is smoothed; a stacked denoising autoencoder is constructed, and the greedy algorithm is used to train the network, and the extracted features are used as the input of deep belief network to train the network; the real-time vibration and noise signals in the machining process are input into the deep network after data processing, and the current surface roughness is set as output by the network.