Search

Cutting tool condition directly affects machining quality and efficiency. In order to avoid severely worn tools used during machining process and fully release the remaining useful life in the meanwhile, a reliable evaluation method of remaining useful life of cutting tools is quite necessary. Due to the variation of cutting conditions, it is a challenge to predict remaining useful life of cutting tools by a unified model. In order to address this issue, this paper proposes a method for predicting the remaining useful life of cutting tools in variable cutting conditions based on Gaussian process regression model incorporated with tool wear mechanism, where the predicted value at adjacent moments is constrained to a linear relationship by the covariance matrix of Gaussian model based on the assumption of progressive tool wear process, so the wear process under continuous changing conditions can be modelled. In addition to that, the input feature space and the output of the model are also enhanced by considering the tool wear mechanism for improving prediction accuracy. Machining experiments are performed to verify the proposed method, and the results show that the proposed could improve the prediction of tool remaining useful life significantly.

Cutting force can provide highly sensitive and quick response to unpredictable variations of the machining system, which has been considered as the most valuable physical signals for machining state monitoring. As a reliable, practical and cost-effective solution for cutting force monitoring, forecasting cutting force using Computer Numerical Control (CNC) inherent monitoring signals has great potential to be applied in real industry. However, existing mechanism-based methods suffer from inaccurate identification process and underlying modeling errors, while data-driven methods are highly data-dependent due to the lack of physical interpretability and generalization ability. This paper proposes a mechanism-based structured deep neural network (MS-DNN) for cutting force forecasting. By taking sub neural networks as approximators for sub models and reserving the connections among the variables of the end-to-end mechanism model without parameter identification, MS-DNN is equivalent to the mechanism model in form, which can naturally inherit the physical interpretability and generalization ability of the mechanism, but is much more powerful in modeling the complex dynamic relationships between cutting force and CNC inherent signals. The proposed MS-DNN is verified on both simulation and real experimental datasets, and the results show MS-DNN can achieve an excellent performance in cutting force forecasting using CNC inherent signals.

Large thin-walled structural parts have been widely used in aircrafts for the purpose of weight reduction. These parts usually contain various thin-walled complex structures with weak local stiffness, which are easy to deform during machining if improper cutting parameters are selected. Thus, local stiffness has to be seriously considered during the machining parameter planning. Existing stiffness calculation methods mainly include mechanics calculation methods, empirical formula methods, finite element methods, and surrogate-based methods. However, due to the structural complexity, these methods are either inaccurate or time consuming. To address this issue, this paper proposes a data-driven method for stiffness prediction of aircraft structural parts. First, machining regions of aircraft structural part finishing are classified into bottom, sidewall, rib and corner to further define the minimum stiffness of machining regions. Then, by representing the part geometry with attribute graph as the input feature, while computing the minimum stiffness using FEM as the output label, stiffness prediction is turned to a graph learning task. Thus, a graph neural network (GNN) is designed and trained to map the attribute graph of a machining region to its minimum stiffness. In the case study, a dataset of aircraft structural parts is used to train four GNN models to predict the minimum stiffness of the defined four types of machining regions. Compared with FEM results, the average percentage errors on the test set are 6.717%, 7.367%, 7.432% and 5.962% respectively. In addition, the data driven model once trained, can greatly reduce the time in predicting the stiffness of a new part compared with FEM, which indicates that the proposed method can meet the engineering requirements in both accuracy and computational efficiency.

The ability of research innovation is particularly important in the era of rapidly changing environment, challenges, and demands, where new knowledge, technologies, and unseen problems are emerging at an unprecedented speed. Engineering college students, especially postgraduate research students are well-trained with domain knowledge and technical skills to solve commonly known problems. However, the lack of a systematic critical thinking framework on research innovation prevents them from generating creative ideas, dealing with changes and new problems, and making step forward and groundbreaking innovations during their research program. This paper presents a seven-question based critical thinking framework, named “IDEA” catechism, which was proposed and implemented as a research training methodology in a research group at a Chinese higher education institution. Through the questions and answers (Q&A) process in the framework, engineering students can be better trained to deeply understand and critically scrutinize their research ideas and methods, and are eventually able to independently develop their own innovative ideas, especially leading them to do original study. The research group has been practicing this framework for several years and succeeded in developing a large number of distinctive innovation with national and international recognition, which were also collected as research cases.

Transfer learning has received much attention recently and has been proven to be effective in a wide range of applications, whereas studies on regression problems are still scarce. In this article, we focus on the transfer learning problem for regression under the situations of conditional shift where the source and target domains share the same marginal distribution while having different conditional probability distributions. We propose a new framework called transfer learning based on fuzzy residual (ResTL) which learns the target model by preserving the distribution properties of the source data in a model-agnostic way. First, we formulate the target model by adding fuzzy residual to a model-agnostic source model and reuse the antecedent parameters of the source fuzzy system. Then two methods for bias computation are provided for different considerations, which refer to two ResTL methods called ResTLLS and ResTLRD. Finally, we conduct a series of experiments both on a toy example and several real-world datasets to verify the effectiveness of the proposed method.

STUDY OBJECTIVES: First, to compare the upper airway's anatomic and aerodynamic characteristics of the edentulous older adults who experience mild, moderate, and severe obstructive sleep apnea (OSA). Second, to examine the correlation between the severity of OSA and the anatomic and aerodynamic characteristic(s) of the upper airway in these edentulous individuals. METHODS: NewTom5G cone beam computed tomography scans of 58 edentulous individuals with mild, moderate, and severe OSA were included in this analysis. 1) Computational models of the upper airway were reconstructed based on cone beam computed tomography images and the anatomical and aerodynamic characteristics of the upper airway were examined by an observer blind to OSA severity. 2) Pearson correlation analysis was used to determine the correlation between apnea-hypopnea index and the anatomic and aerodynamic characteristics of the upper airway. RESULTS: Compared with edentulous patients with mild and moderate OSA, those with severe OSA have a more hourglass-shaped upper airway. The severity of OSA, namely, apnea-hypopnea index, was significantly correlated with the length, shape, and minimum cross-sectional area of the upper airway. During inspiration, the mean velocity of the airflow within the upper airway of the edentulous patients with severe OSA was higher than that of patients with mild and moderate OSA. During both inspiration and expiration, apnea-hypopnea index was found to be significantly correlated with maximum velocity (P = .05) and airway resistance (P = .024, 0.038). CONCLUSIONS: The edentulous patients with severe OSA have a more hourglass-shaped upper airway. The findings also suggest that, during inspiration, the airflow travels faster in edentulous patients with severe OSA than in those with mild or moderate OSA. CLINICAL TRIAL REGISTRATION: Registry: ClinicalTrials.gov; Name: The Effect of Nocturnal Wear of Dentures on Sleep and Oral Health Related Quality of Life; URL: https://clinicaltrials.gov/ct2/show/NCT01868295; Identifier: NCT01868295. CITATION: Chen H, Elham E, Li Y, et al. Comparison of anatomic and aerodynamic characteristics of the upper airway among edentulous mild, moderate, and severe obstructive sleep apnea in older adults. J Clin Sleep Med. 2022;18(3):759-768.

With the increase of power load, the overheating problem of oil-immersed transformer in operation cannot be ignored. Therefore, it is necessary to propose reasonable heat dissipation enhancement measure for oil-immersed transformer in operation and study the influence of the measure on the internal temperature of transformer. The heat transfer process of oil-immersed transformer is analyzed, and the air-side heat transfer coefficient is pointed out to be the important parameter limiting the heat dissipation capacity of the transformer. In order to improve heat dissipation capacity, the measures based on cooling fans and falling film are proposed and the calculation methods for the air-side heat transfer coefficient are conducted, respectively. Moreover, the top-oil temperature and hot-spot temperature under two measures are compared by the thermal-fluid coupling model of oil-immersed transformer. For the transformer operating at rated load and the ambient temperature of 25?C, after the cooling fans are adopted, the top-oil temperature and hot-spot temperature can be reduced by 22.4?C and 20.7?C, while after the falling film is adopted, the top-oil temperature and hot-spot temperature can be reduced by 31.2?C and 28.7?C, respectively. The results show that the falling film is a more effective heat dissipation enhancement measure, which can meet the higher load demand of oil-immersed transformer.

Accurate cutting force prediction serves as an important reference to the optimization of numerically controlled machining process. Traditional cutting force modeling via theoretical cutting mechanism hampers accurate prediction for actual machining process due to its highly suppressed modeling flexibility. On the other hand, machine learning based modeling approaches demand large amount of diversified labeled samples to achieve comparable prediction results, while collecting these samples can be tedious and costly because the cutter workpiece engagement (CWE) keeps changing during actual process. This paper presents a cutting force prediction model, named ForceNet, which incorporates elementary physical priori into structured neural networks to predict cutting force for end-milling process of complex CWE. The main idea is to use grayscale images to represent CWE geometry, providing a universal input to the ForceNet. Unlike traditional deep neural networks served as an unexplainable black box, the core of the ForceNet is constructed by the vector summation of directional primitive cutting force elements, which are approximated using elementary neural networks. Preliminary results indicate that ForceNet outperformed existing methods not only with greater prediction accuracy in unseen cutting situations, but also with less training data needed thanks to its inherent neuro-physical structure.

Aircraft structural parts are important and high-value parts used to constitute the frame of the aircraft, and are usually produced by NC machining, where the machining parameters are significant for the machining quality, efficiency, and cost. In the process planning, there are hundreds or even thousands of machining operations that require separate machining parameters, which is a huge task for the existing optimization-based methods that rely on iterative optimizations. Due to the complex structures and high requirements, the existing expert system-based methods require plenty of additional modifications. Recently, with the development of artificial intelligence, data-driven methods are used in machining parameter planning, which mines the knowledge and rules hidden in the historical data. However, the existing data-driven models require a large amount of training data and lack interpretability. To address this issue, this paper proposes an informed machine learning method for machining parameter planning, which introduces multiple prior constraints into the data-driven model. First, the part model is represented as an attribute graph, and the cutting area of each machining operation is correlated to a subgraph, which is used to obtain the vectorized representation of machining operation that covers cutting area and process information. Then, by fitting the mapping between the vectorized machining operation and the machining parameters, the knowledge and rules are learned. Next, to introduce prior constraints into the data-driven model, the constraint loss is designed and incorporated into the original loss function. The proposed method can generate machining parameters for all the machining operations in batch, thereby greatly reducing the human interactions. In the case study, the historical processing files of aircraft structural parts are used to train the proposed model for planning cutting width, cutting depth, spindle speed, and machining feedrate. The results show that the demand for training data is reduced and the prediction accuracy is improved with prior constraints.

Data-driven modelling has shown promising potential in many industrial applications, while the expensive and time-consuming labelling of experimental and simulation data restricts its further development. Preparing a more informative but smaller dataset to reduce labelling efforts has been a vital research problem. Although existing techniques can assess the value of individual data samples, how to represent the value of a sample set remains an open problem. In this research, the aggregation value is defined using a novel representation for the value of a sample set by modelling the invisible redundant information as the overlaps of neighbouring values. The sampling problem is hence converted to the maximisation of the submodular function over the aggregation value. The comprehensive analysis of several manufacturing datasets demonstrates that the proposed method can provide sample sets with superior and stable performance compared with state-of-the-art methods. The research outcome also indicates its appealing potential to reduce labelling efforts for more data-scarcity scenarios.