Recently, owing to the effects of the global coronavirus pandemic and decline in the working-age population in some countries, production efficiency and labor productivity are being actively improved at manufacturing and production sites by introducing digital technologies. Digital transformation (DX), Internet of Things (IoT) utilization, digital twin application, etc., are the representative efforts being made in the manufacturing and production fields these days [1–4].
Given these trends, to improve the labor productivity in companies, especially local small and medium-sized manufacturing companies, through DX and IoT, we proposed a failure detection and failure prediction maintenance technology for hydraulic pressing machines used for bending steel materials, such as iron plates. Some parts of the hydraulic pressing machine need long time to repaired, thereby resulting in long-term machinery outage and suspension of production activities. Therefore, we aimed to realize better planned production activities to improve labor productivity by diagnosing and predicting machine failures and abnormalities at the earliest.
Many vibration-based systems for detecting machine failures and abnormalities have been devised and proposed.
In particular, many attempts have been made to clarify the state of rotating machines by vibration measurement. These are summarized in detail by Tiboni et al [5], Kirankumar et al [6] and Malla and Panigrahi [7].
With regard to rolling bearing, Pandey et al [8] focused on the mode shape of a curved surface and carried out damage detection, and the review by Tandon and Choudhury [9], which deals with vibroacoustic measurement techniques for rolling elements, are well known. In addition, Kozochkin et al. studied the diagnosis using vibration measurement in the cutting process [10, 11] and Sabirov et al. studies the method for the end mill diagnosis [12].Further, Qiu et al. investigated the diagnosis method using Wavelet filtering [13], Ban et al. studied sounds generated from roller bearings for the diagnosis [14], Lei et al. studied the diagnosis method using Kurtogram Method [15], Randall showed the tutorial of the vibration classification method [16], Zhao et al. [17], and Immovilli et al. [18] investigated the relationship between rotational speed and shaft radial load and vibration in the bearing system.
Research other than rolling bearing has also been extensively conducted. There have been studies on fault diagnosis for gearboxes [18], rotating blades [19], wind turbines [20–22], motors [23], pumps [24], and fans [25]. Studies on non-linear systems is also underway, and research on diagnosis by monitoring space topology changes have been studying [26, 27]. In recent years, research using machine learning by Support Vector Machine (SVM) and k-nearest neighbor (kNN) has also been widely promoted [28]. This research is currently the most active. Sharma et al. [29] used SVM and artificial neural network (ANN), and Liontos and Georgiou used ANN combined with Proper Orthogonal Decomposition (POD) [30].
In this way, studies on fault diagnosis of machines has been conducted all over the world. However, there are few cases related to hydraulic presses and few development examples related to systems that detect machine failures and abnormalities that involve intermittent operation by manual operate, e.g., hydraulic cylinders of pressing machines (most are application examples of machines that perform continuous operation). Many studies have been conducted to detect abnormalities in engine cylinders by vibration measurement [e.g., 31], but there are no examples of research on hydraulic press cylinders. Furthermore, although there were many cases where these systems were introduced from the beginning or manufacturing stage of the machinery, few systems were introduced in machinery, as targeted in this research, that has been used for more than a decade by the manufactures, which is considered to be the lack of know-how for introducing the system in the form of retrofitting. We have been conducting research for the following two objectives: (1) to construct a failure and abnormality detection system for machines with intermittent operation or manual operation, and (2) to construct a failure and abnormality detection system that can be retrofitted to existing machinery.
When detecting machine failures and abnormalities, it is desirable to measure the physical quantity that is correlated with the phenomenon that causes the failure and abnormality. For example, it is possible to detect failures and abnormalities more accurately by installing a temperature sensor in the case of failures and abnormalities caused by heat generation and a strain sensor in the case of strain. However, it is difficult to install the desired sensor at the desired position and guarantee the measurement accuracy when constructing a system that detects failures and abnormalities after a machinery is installed. Additionally, sensors require periodic calibration and replacement poses a problem from the viewpoint of continuous usability, strength, and durability. Therefore, to address these problems, we used a sensor with excellent continuous usability, strength, and durability to measure vibrational acceleration. The proposed sensor can be easily installed afterwards.
Many techniques and useful methods have been proposed for detecting failures and abnormalities in machines that operate continuously, e.g., machines that continue to operate at a constant rotation speed or cycle. On the other hand, in machines, such as hydraulic presses, where manual operation are performed intermittently by workers and operators, methods for detecting failures and abnormalities have not been studied much. Moreover, it is assumed that in many cases, data acquirement, data evaluation, and failure and abnormality detection is performed in a state where ad hoc measured and accuracy are not guaranteed.
In this research, we created a regression model using the Gaussian process regression model, which is a non-parametric stochastic model, for the acquired data to grasp the state of the machine with certain degree of accuracy even when the machine is used in uncertain intermittent operation and manual operation to devise an algorithm for judging failures and abnormalities.