are cp components in the code layer, then the SF set SFAEN can be defined as {h1, h2, …hcp}. This feature extraction method is very similar to adopting the other well‐known dimensionality reduction technique: principle component analysis (PCA).
2.4 Case Studies
Four practical examples using real‐world data are respectively demonstrated to validate techniques of data acquisition and data preprocessing addressed in the previous sections. Details are described as below.
2.4.1 Detrending of the Thermal Effect in Strain Gauge Data
To detect force and torque during machining, a smart tool holder is developed and used in an CNC milling machine. When several corresponding gauges are attached to the holder, a strain gauge can be used to detect variation in the bending and torsion of the tool holder based on the proportional ratio of the resistance to the length of the stain gauge. As illustrated in Figure 2.23, the values of strain on the tool holder are sensed and detected by Wheatstone bridges, digitized using an ADC, processed via a microprocessor, and transmitted to an edge computer by using the message queuing telemetry transport (MQTT) protocol through a Wi‐Fi module.
Figure 2.23 Using a smart tool holder to detect tool state.
The edge computer located near the CNC machine receives and processes strain values and issues tool events to the controller when tool breakage is detected or tool’s RUL is short. A tool holder is stiff enough to enable clamping of a tool under various machining conditions and lead to tiny machining variation in the length and resistance of a strain gauge. Although a high‐gauge‐factor sensor is employed, a length difference (<1 μm) in a tool holder can be detected during machining. However, the strain gauge appears to have considerable thermal variations even in a stationary state. Thus, one challenge is how the thermal effect in strain‐gauge data can be removed to derive effective strain values; the details are described in [17] via IEEE DataPort.
For example, Figure 2.24a depicts raw signals collected during machining for 2.8 s (sampling rate of 10 kHz). Because of the high heat capacity of a real machine, the thermal variation can be assumed to be constant within a short period such as 5 s. After applying the wavelet de‐noising method with five levels (denoted DB5), the thermal trend can be derived as illustrated in Figure 2.24b. Then, the de‐trended data can be obtained by subtracting the thermal trend from the raw data; the result is depicted in Figure 2.24c.
Figure 2.24 Detrending of the thermal effect in strain‐gauge data: (a) before detrending; (b) the thermal trend; (c) after detrending.
The thermal trend in the raw signal can be removed and normalized to the same criterion by subtracting the mean value. As shown in Figure 2.25, the darker and lighter lines represent the signals before and after de‐noising using the wavelet mother function DB3, respectively. The signals collected during machining imply that the signals contain the dry‐run and tool‐use periods.
Figure 2.25 De‐noising signals to highlight differences between dry‐run and tool‐use periods.
By comparing with the values measured during the dry‐run period (range ± 0.035 mV), the values measured during tool‐use can be filtered from the background noise and derived in the range of ±0.083 mV. After de‐trending and de‐noising the raw data, the ratio between the dry‐run and tool‐use signals can be improved from 1 to 2.37, which makes the extraction of effective features for modeling considerably easier.
2.4.2 Automated Segmentation of Signal Data
To detect tool wear during machining, a vibration‐based evaluation method is developed and used on a CNC milling machine. During any machining operation, a tool holder chunked by the spindle holds the cutting tool in place as precisely and firmly as possible. The spindle stability affects the quality of tool holder and cutting tool. Thus, a high‐resolution accelerometer attached close to the spindle is adopted to monitor the cutting‐tool wear based on the changes of spindle vibration.
However, when M codes are not available in some cases, for example, the required number of DIO is too large to achieve some complicated operations, so that the raw data cannot be timely divided into several critical parts of final machining process by M Codes during the machining time. When dealing with the segmentation issue under the condition of an insufficient number of M codes, the feasible solution is to decrease the usage of the pairs of M codes by extending the duration of each data acquisition. In this way, however, not only the specified final machining parts that affect product quality most are included but various types of machining operations would be involved during each data acquisition.
In this manner, the acquired vibration data may be a long signal that contains the process during the idling (dry‐run) and the real machining (tool‐workpiece contact) periods. Thus, the challenge is how to automatically segment the collected data so as to estimate the tool‐wear status.
Three‐axis vibration data sampled at 2,048 Hz during the drilling operation of seven holes on a medal plate with material FDAC are respectively illustrated in Figure 2.26. To increase tool availability, a 0.5 mm hole is pecked by a two‐flute tool in each cycle until a total depth of 4 mm is achieved. The feed rate and spindle speed are 100 mm/min and 4,500 rpm, respectively. Obviously, the main loading happens in the Z‐axis vibration, and the bottom of Figure 2.26 shows real machining periods of seven parts, which are manually segmented and numbered from 1 to 7. The details of this application case can also be found in [18] via IEEE DataPort.
Figure 2.26 Collected vibration signals (including idling and machining periods).
To automatically segment the aforementioned Z‐axis data for identifying the actual drilling periods, an AEN model integrated with an encoder, code, and decoder is used to learn the idling characteristics under specific conditions. In this segmentation case, the encoder (four‐layer structure with 32, 16, 8, and 4 nodes, respectively) compresses the inputs into code in the middle layer, and the decoder (the inverse structure with 4, 8, 16, and 32 nodes) decompresses code into the outputs. Only one node in the code layer is used to evaluate
