Condition monitoring of deep drilling process for cooling channel making in hot press die
Deep drilling operation is one of the major process that is widely used in the manufacturing industry. To make a cooling channel of hot press forming die, deep drilling is a crucial process which is the drilling depth is 10 times of the drill bit diameters itself. However, the major complications th...
| Summary: | Deep drilling operation is one of the major process that is widely used in the manufacturing industry. To make a cooling channel of hot press forming die, deep drilling is a crucial process which is the drilling depth is 10 times of the drill bit diameters itself. However, the major complications that occur is the drill bits will become wear and breaks as the drilling depth keep increasing. This will impact the efficiency of the drilling process. To overcome this drawback, Tool Condition Monitoring (TCM) was introduced to refining the quality of the drilling process by monitor the drilling operation, whether by direct or indirect monitoring, thus will improve tool life expectancy by notifying the operator to stop the machine. SKD 61 which is widely used as die material and High Speed Steel (HSS) drill bit was chosen. To improve accuracy, Tri-axial Accelerometer (PCB356B21) was used to detect the vibration of the drill bit when drilling process occurs. The data obtained from this experiment is in the form of acceleration and Fast Fourier Transform (FFT) signal which is acceleration x, acceleration y, acceleration z, FFT x, FFT y, and FFT z. Tool condition can be classify into five types by using this data which is good condition, small corner wear, medium corner wear, large corner wear and fracture. To classify this data, machine learning method such as Support Vector Machine (SVM) and Artificial Neural Network (ANN) was employed. SVM performs classification process based on the data input vector that comprise as fault in the machine. The fault is then being produced as the pattern and SVM will recognize thus classify this pattern corresponding to the fault. Nevertheless, the major downside of SVM is less accurate of classifying result due to the data over-fitting. To get an accurate result, the data is compared with Artificial Neural Network machine learning. ANN performs an excellent classifying by determining the correct set of input data, number of hidden layers, target data, and applying a suitable algorithm, which is proven better classifying result than SVM in the aspect of classifying accuracy and number of errors. Consequently, ANN is the most suitable method to classify tool condition, and are capable to be employed for online tool failure detection system which is beneficial for optimizing tool condition in industries. |
|---|