Wear Monitoring of Single-Point Dresser in Dry Dressing Operation Based on Neural Models

Abstract

The monitoring of different machining processes has been studied for years, however many processes still do not have a final solution for their controls. The dressing, as it is of great importance in the finishing of workpieces produced through the grinding, is an operation whose monitoring becomes necessary. In order to make the dressing automation and, in this case, the process of dresser exchange, there is a need for efficient and lowcost monitoring. The vibration sensor has great potential, but it is still little used for this purpose. In this work the vibration sensor and neural models were used to classify the wear of dressing tools for three different conditions. Dry dressing tests and data acquisition were performed in a surface-grinding machine. The raw signals were further filtered in different frequency bands. Then, two statistics were computed, which served as inputs to the neural models. The results were quite satisfactory for some models.

Dressing Tool Condition Monitoring through Impedance-Based Sensors: Part 2—Neural Networks and K-Nearest Neighbor Classifier Approach

“…The CVD diamond tip presented less than 10% of worn area for 600 passes due its very high wear resistance (loss of 0.06 mm 2 ). This result is in agreement with previous research work, such as [37,40], where authors addressed wear condition assessment of single-point dresser by studying the frequency content of acoustic emission and vibration signals for several frequency bands. Neural network models were applied to classify the diamond wear during dressing with satisfactory results although high standard deviations were observed, probably due to different friability properties of the utilized diamonds.…”

“…In Figure 6b (model#3), it is possible to see that one feature belonging to the D1 damage class was classified as D3, as well as some features belonging to D3 were classified as D2. This result again agrees with the literature, since most of the damage features erroneously classified occurred in regions close to the borders, as reported in [37] where the authors did not consider them as serious errors because these boundaries can be difficult to determine.…”

“…Brzezinski et al [35] proposed a new method for grinder dressing fault mitigation using real-time peak detection. In Junior et al [36,37], the evaluation of single-point dresser wear was proposed through intelligent spectral analysis of vibration signals. Miranda et al [38] presented a methodology to estimate the wear of single-point dressers based on AE, vibration and fuzzy models.…”

“…In addition, a study was conducted to identify the best ANN architecture. Each ANN model was trained and tested by combining the input patterns of two consecutive frequency sub-ranges, based on [37,40], to yield a combined input pattern containing 12,000 features for each of the three resulting combined frequency sub-ranges: [0–20 kHz + 20–40 kHz], [40–60 kHz + 60–80 kHz], [80–100 kHz + 100–120 kHz] (Table 3). The scope was to find the best frequency sub-range combination capable to provide the reliable monitoring of dressing tool condition.…”

A New Approach For Dressing Operation Monitoring Using Voltage Signals Via Impedance-Based Structural Health Monitoring

“…Tradicionalmente, o monitoramento da condição da ferramenta de dressagem é feito pelo operador, porém há um grande interesse de pesquisadores em desenvolver tecnologias que possam monitorá-la automaticamente, contribuindo efetivamente para otimização do processo de retificação ( Junior et al, 2017).…”