A Novel Local Time-Frequency Domain Feature Extraction Method for Tool Condition Monitoring Using S-Transform and Genetic Algorithm

Rad, Javad Soltani; Zhang, Youmin; Chen, Chevy

doi:10.3182/20140824-6-za-1003.02515

Cited by 5 publications

(2 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since the data were collected from a real milling machine (instead of simulation models or experimental platforms) working under operational conditions close to industrial applications, it provides important information for researchers to study the relationship between health states and measurements. 9,34,35 In addition, we propose to use the leave-one-out cross-validation method to cope with the overfitting problem caused by the limitation regarding the number of samples. By doing so, we deem the milling data set is sufficient for an effective study as conducted in this paper.…”

Section: Discussion and Resultsmentioning

confidence: 99%

“…A variety of typical monitoring indices can be obtained from the time-domain and frequency-domain features. 7 Advanced signal processing techniques such as mel-frequency cepstral coefficient 8 and S-transform 9 have also been employed by some researchers to extract features that represent the characteristics of the cutting process (information) and to separate these features from various noise disturbances. However, features extracted from an independent data source are not equally informative: certain features may correspond to noise, not information; others may be correlated or not relevant to the target.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Identifying optimal features for cutting tool condition monitoring using recurrent neural networks

Mechefske

Kim

2014

Advances in Mechanical Engineering

View full text Add to dashboard Cite

Identification of optimal features is necessary for the decision-making models such as the artificial neural network to achieve effective and robust on-line monitoring of cutting tool condition. Most feature selection strategies proposed in the literature are for pattern recognition or classification problems, and not suitable for prognostic problems. This paper applies three parameter suitability metrics introduced in previous similar studies for failure-time analysis and modifies them for failure-process analysis which allows for the unit-wise variation of the component in a population. The suitability of a feature used for cutting tool condition monitoring is determined by its fitness value calculated based on the three metrics. Two types of recurrent neural network are employed to analyze the prognostics ability of the features extracted from multi-sensor signals (acoustics emission, motor current, and vibration) collected from a milling machine under various operating conditions. The analysis results validate that the fitness value of a feature can depict its prognostic ability. It is found that adding more features which share abundant information does not increase the prediction performance but increases the burden on the decision-marking models. In addition, adding features with low fitness values may even deteriorate the prediction.

show abstract

Section: Discussion and Resultsmentioning

confidence: 99%