cutting forces, component wears, thermal variation, machine structures, or tool breakages) and released in the form of the acoustic radiation into an electric signal. These materials’ voice can be received and interpreted by the AE sensor.
The most widely used AE sensor is the piezoelectric transducer that converts the mechanical energy into an electrical voltage signal. The AE sensor has the highest sensitivity in frequency response from 30 k Hz to 1 M Hz, which is significantly higher than that of microphone (20–100 k Hz), accelerometer, or strain gauge. Thus, this advantage makes the obtained signals less likely to be disturbed or attenuated by the mechanical structures or components. Figure 2.10 demonstrates the installation of an AE sensor on a tool and next to a strain gauge.
Figure 2.10 Installation of an AE sensor.
However, to acquire such high‐frequency data, a relatively high corresponding sampling rate is required. Therefore, a sensing system with large storage and strong computing power for the large volumes of data is indispensable. In addition, the high sampling rate also increases the difficulty of data preprocessing for raw AE signals.
Sensor Fusion
There is not a single sensor that can perfectly capture all signs released from the equipment during production. Every type of sensors has its own limitations in different aspects; therefore, it is difficult to obtain a comprehensive result by using only one sensory source.
To meet the requirement of enhanced accuracy with greater robustness under the varying environment, the sensor fusion technique that combines sensory data from individual and multiple sensory sources in a complementary manner is extensively proposed to improve the resolution and reliability of the acquired signals. Strictly speaking, multiple sensors with noncomplementary measurements and various features extracted from single signal can only be regarded as a multi‐sensor but not a sensor fusion system, since too much redundant information may decrease the accuracy [6, 7].
Note that, the sensor accuracy might be worsened with the increase of the distance between the sensor and the cutting zone. If the direct sensors cannot be attached on the surface closer to the cutting zone, the sensor fusion system can solve this problem by attaching two closely related sensors such as the combination of the accelerometer/current sensor, or the accelerometer/dynamometer, to provide the cross‐validation scheme between dependent signals without any loss of important information.
Generally speaking, AE sensors have very sensitive frequency and transient response, which are very adequate for combining with the current sensor, accelerometer, or dynamometer. Figure 2.11 illustrates a machinery spindle monitored by a sensor fusion system using an accelerometer and one thermal couple.
Figure 2.11 Sensor fusion system comprising an accelerometer and a thermal couple.
Figure 2.12 demonstrates a sensor fusion system with five built‐in sensors near the cutting zone on a computer numerical control (CNC) machine to capture critical information. The accelerometer and the thermal couple are mounted on the spindle, CT is attached on the power cable, AE is mounted and dynamometer is fixed on the machine table.
Figure 2.12 Sensor fusion system using five types of sensors.
2.2.1.2 Manufacturing Parameters Acquisition
Manufacturing parameters are actual values retrieved from specific devices (e.g. machine controllers) during the production process. The actual values of manufacturing parameters are based on manually set target values and usually affected by numerous variables in the real‐world environment.
For example, two commonly measured parameters that affect the injection molding process are the actual pressure value and temperature value. The manufacturing parameters are periodically updated under the supported refreshing rate of the machine controller.
Nowadays, with many easy‐to‐use Ethernet‐based communication standards for industrial environments, such as Ethernet for Control Automation Technology (EtherCAT), MTConnect, and Open Platform Communication Unified Architecture (OPC‐UA), accessing and retrieving dynamic manufacturing parameters seem not as difficult compared to installing external sensors. This advantage enables manufacturing parameters to be widely applied to record stable data that is not likely to change, so that lower sampling rate can suffice the demand (e.g. temperature, tool number, feed rate, etc).
However, high‐frequency data are not available via applying communication standards. The machine controller is designed to continuously guarantee the machining quality of all time and it executes the machining processes with a high computational complexity. Since frequent requests for accessing parameters have a high risk to interrupt normal operations, the sampling rate is usually set to be within 300–500 μs for each cycle. Once the machine controller overloads, returned values of process parameters might be just default values rather than actual values. Thus, the mainstream methods of machining condition monitoring usually adopt the combination of both sensor signals and process parameters. More details of communication standards can be found in Chapter 3.
2.2.2 Metrology Data Acquisition
The metrology data of the product quality aim to conform to customer requirements and minimize defect costs via quality control. Product quality means the overall quality of a product, which is what really matters for the customers.
A quality report that records measurement results for a specific product (e.g. dimensional accuracy, surface roughness, and position tolerance, etc.) can be completed via the inspection of various measurement equipment, such as the coordinate measuring machine (CMM) or the automated optical inspection (AOI) device, so that it shows whether the status of the product conforms to customer's requirements, specifications, and expectations. For any case, the best situation is to measure the practical quality of all end‐products.
Note that, metrology delay varying with the inspection frequency and measurement place would restrict the feasibility of total inspection. Therefore, intelligent applications [1–8] should come in place to solve this issue.
In short, process data are the major contributors to metrology data. They are used as the predictor variables; while the metrology data of products can be used as a label variable or category for typical supervised learning or reinforcement learning methods. In other words, given that the data acquisition quality is reliable, the causation between process data and metrology data would be a strong correlation that efficiently indicates if the manufacturing process performs well or not.
Thus, the root cause analysis can be conducted to narrow down to the specific process steps that affect product quality. Manufactures, therefore, have a chance to make their production line free from deficiencies or defects via continuous improvement and corrective actions.
2.3