1. Ph.D Theses
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/1/11
Browse
2 results
Search Results
Item Falut Diagnosis of Single Point Cutting Tool Through Online and Offline Monitoring Techniques(National Institute of Technology Karnataka, Surathkal, 2016) N, Gangadhar.; S, Narendranath; Kumar, HemanthaTool condition monitoring plays a crucial role in automated industry to monitor the state of cutting tool. It prevents any hazards occurring to the machine, avoid deterioration of the surface finish on end product and it helps to introduce a new tool in an instant at which the existing tool has worn out toensure safety, productivity and optimum performance of the metal cutting process. In the present research work,fault diagnosis of single point cutting tool is investigated based on the vibration signals and cutting force signals on an engine lathe. Vibration signals and cutting force signals corresponding to a healthy insert (baseline) anddifferent types of industrial practical worn out insertswere recorded. The research work is carried out in three phases. The first phase investigates fault diagnosis of cutting tool using signal processing techniquessuch astime domain, spectrum, cepstrum, continuous wavelet transform (CWT), recurrence plots (RPs) and recurrence quantification analysis (RQA). The result shows that recurrence plots and recurrence quantification analysis were useful for revealing post fault detection and diagnosis of worn states of the inserts. The second phase of research workpresents fault diagnosis of cutting tool using machine learning approach based on vibration signals. From the vibration signals, statisticalfeatures, histogram features, discrete wavelet transform (DWT) features and empirical mode decomposition (EMD) featureswere extracted. Principle component analysis (PCA) and J48 algorithm (decision tree) were used for important feature selection/reduction. Artificial neural network (ANN), Naïve Bayes, Bayes net, support vector machine (SVM), K-star and J48 algorithm classifiers have been used to classify the different fault conditions. Classification accuracy is found to be reasonably good with J48 algorithm feature selection compared to PCA. The third phase presents the results of investigations undertaken to find suitability of vibration signals and cutting forces to detect the condition of tungsten carbide cutting tool insert, surface roughness and type of chip formation. The results show that there is an increase in the level of acceleration and cutting force at faulty tool condition ascompared with the healthy condition of the tool. Based on this finding, cutting tool acceleration and cutting forces can be used to predict the cutting tool condition, surface roughness and chip formation type. Qualitative comparisons of the computational predicted forces are drawn by plotting the trends of the predicted forces together with the measured forces. The Deform-3D has correctly predicted this trend which is consistent with the experimental trends of the cutting forces components. Tool wear analysis has been carried out on the worn tungsten carbide insert cutting tools to find the tool wear mechanisms.Based on SEM micrographsof worn surface of the cutting tool,micro-abrasion, micro-attrition, adhesion and micro-fatigue behaviors are identified as the dominant kinds of wear mechanisms.Item Condition Monitoring of Face Milling Tool Using Vibration and Sound Signals(National Institute of Technology Karnataka, Surathkal, 2018) C. K, Madhusudana; Kumar, Hemantha; S, NarendranathFault diagnosis of the cutting tool is very essential for improving the quality and maintaining the accurate dimension of the products during machining process. The milling is a multi-toothed metal removing process. In face milling, because of dynamic variation of cutting forces, thermo-mechanical shocks and vibration, which results in catastrophic tool failure along with gradual wear of the tool inserts. Wear development during machining can reach up to unacceptable level, resulting in inaccurate dimension and poor surface finish of the components. Monitoring the condition of the cutting tool during face milling operation is a vital role before the tool causes any damage on the machined surface which becomes highly valuable in order to avoid loss of products, damage to the machine tool and associated loss in productivity. Keeping in view of the automation, it is necessary to choose an effective and efficient method for monitoring the cutting tool condition without affecting the machining setup and the work material. This study mainly deals with the fault diagnosis of the face milling tool using vibration and sound signals through signal processing techniques and machine learning approach. The face milling is a machining process with an intermittent cutting action. The milling tool will undergo different types of faults such as flank wear, breakage and chipping which occurs predominantly during milling. The vibration and sound signals under these faulty and healthy milling tool conditions are acquired and these signals are further analyzed. Current research work is mainly categorized into two phases. The first phase is to detect/diagnose the face milling tool conditions by analyzing the vibration and sound signals using signal processing techniques. The signal processing techniques such as time-domain analysis, spectrum analysis, cepstrum analysis and continuous wavelet transform (CWT) method are applied to recognize the face milling tool conditions. The cepstrum analysis has been applied for the first time in fault detection of the face milling tool and has provided the sufficient information about the face milling tool condition using both vibration and sound signals. Generally conventional data processing is computed in time or frequency domain which is not suitable for analyzing non-stationary signals. In order to overcome the lack of a global view on how to develop machining monitoring systems based on artificial intelligent models, machine learning approach is one of the best methods for developing an effective tool condition monitoring (TCM) system.In the second phase, fault diagnosis studies of the face milling tool using vibration and sound signals based on artificial intelligence techniques are conducted. Fault diagnosis of the different tool conditions based on machine learning technique is basically comprised of three steps; feature extraction, feature selection and feature classification. Different features such as, statistical features, histogram features, discrete wavelet transform (DWT) features and empirical mode decomposition (EMD) features are extracted from the acquired vibration and sound signals. For example, features such as skewness, mode, standard error, maximum, minimum, range, sum, mean, standard deviation, median, sample variance and kurtosis are computed from each acquired vibration and sound signals will serve as statistical features. The important features out of all extracted features are to be selected using induction based on decision tree technique (ID3 algorithm or J48 algorithm). The artificial intelligence techniques such as support vector machine (SVM), Naïve Bayes algorithm, artificial neural network (ANN), decision tree algorithm and K-star algorithm are used to classify the data using selected features. Fault diagnosis analysis with acquired vibration and sound signals are carried out by making use of different combinations of feature extraction methods and different classifiers with selected features based on decision tree technique. Overall results have shown that the vibration signal based fault diagnosis has given better classification accuracy than the sound signal based fault diagnosis. The current research work has demonstrated that the statistical features served as best features among all other features extracted such as, EMD features, Histogram features and DWT features. It is also found that the Naïve Bayes algorithm provides best classification accuracy in comparison with other classifiers used such as SVM, ANN, decision tree and K-star algorithm. Based on research work, it is proposed that the combination of statistical features and the Naïve Bayes algorithm as classifier is the best feature-classifier pair using vibration signals in tool condition monitoring system for the face milling process.