An Acoustical Machine Learning Approach to Determine Abrasive Belt Wear of Wide Belt Sanders Maximilian Bundscherery Thomas H. Schmitty Sebastian Bayerly Thomas Auerbachzand Tobias Bocklety

2025-04-30 0 0 2.29MB 4 页 10玖币

侵权投诉

An Acoustical Machine Learning Approach to

Determine Abrasive Belt Wear of Wide Belt Sanders

Maximilian Bundscherer ∗†, Thomas H. Schmitt †, Sebastian Bayerl †, Thomas Auerbach‡and Tobias Bocklet†,

†Department of Computer Science -Technische Hochschule N¨

urnberg Georg Simon Ohm N¨

urnberg, Germany

‡Production Technology -Hans Weber Maschinenfabrik GmbH Kronach, Germany

∗Email: maximilian.bundscherer@th-nuernberg.de

Abstract—This paper describes a machine learning approach

to determine the abrasive belt wear of wide belt sanders used

in industrial processes based on acoustic data, regardless of the

sanding process-related parameters, Feed speed, Grit Size, and

Type of material. Our approach utilizes Decision Tree, Random

Forest, k-nearest Neighbors, and Neural network Classiﬁers to

detect the belt wear from Spectrograms, Mel Spectrograms,

MFCC, IMFCC, and LFCC, yielding an accuracy of up to 86.1%

on ﬁve levels of belt wear. A 96% accuracy could be achieved with

different Decision Tree Classiﬁers specialized in different sanding

parameter conﬁgurations. The classiﬁers could also determine

with an accuracy of 97% if the machine is currently sanding

or is idle and with an accuracy of 98.4% and 98.8% detect the

sanding parameters Feed speed and Grit Size. We can show that

low-dimensional mappings of high-dimensional features can be

used to visualize belt wear and sanding parameters meaningfully.

Index Terms—acoustic sensors, abrasive belt wear, tool wear,

machine learning, industrial process, wide belt sanding machines

I. INTRODUCTION

Wide belt sanding machines are commonly used in indus-

trial processes to remove material from a workpiece by an

abrasive belt [1]. This paper uses acoustic data and machine

learning methods to determine the abrasive belt wear of a

WEBER KSF 1350 wide belt sander. Our models aim to

optimize tool change intervals, a critical optimization criterion

in industrial processes [2]. We used lightweight approaches

that potentially allow our models to be deployed on low-

cost endpoints such as a Raspberry Pi or ESP32. The use of

acoustic sensors, like microphones as additional sensors and

low-cost edge devices for classiﬁcation in industrial processes,

is convenient and inexpensive since no machines or production

chains need to be strongly modiﬁed or rebuilt. Fig. 1 gives a

schematic overview of our approach.

A. Our contributions

•Recording and labeling of wide belt sanding machine

operations with 18 sanding process-related parameter

conﬁgurations.

•Describing and evaluating a machine learning based

method for classifying ﬁve levels of belt wear, machine

state (is actual sanding or is idle), and sanding parameters.

•Meaningful low-dimensional mapping of high-

dimensional features.

This study is partially supported by the European Social Funds (ESF)

No. R.6-V0332.2.43/1/5.

Fig. 1: Schematic overview of our approach.

•Models that use acoustic data only and can operate on

lightweight edge devices.

B. Related work

In the literature, a distinction is made between the terms

grinding, sanding, and polishing to describe abrasive pro-

cesses. [1] presents an approach to detect abrasive grinding

belt wear over three levels on a grinding machine using sound

and current as features under varying grinding parameters.

Convolutional Neural Networks are used to predict belt grind-

ing tool wear in a polishing process using 3-axes force and

vibration data as models input by [3]. Other publications

have focused on details of the grinding process [4] and tool

condition monitoring [5]. This study’s novelty is applying

acoustic machine learning detection to wide belt sanding

machines and focuses on the varying wood sanding parameters

Feed speed, Grit Size, and Type of material.

II. DATA

Wide belt sanding machines have sanding process-related

parameters that inﬂuence the acoustics characteristics of the

process more than the abrasive belt wear itself. This phe-

nomenon has also been observed in grinding machines [6].

The sander was allowed to cool down between recordings to

mitigate the inﬂuence of different engine temperatures.

arXiv:2210.13273v1 [eess.AS] 24 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

AnAcousticalMachineLearningApproachtoDetermineAbrasiveBeltWearofWideBeltSandersMaximilianBundscherery,ThomasH.Schmitty,SebastianBayerly,ThomasAuerbachzandTobiasBocklety,yDepartmentofComputerScience-TechnischeHochschuleN¨urnbergGeorgSimonOhmN¨urnberg,GermanyzProductionTechnology-HansWeberMaschinenfa...

展开>> 收起<<

An Acoustical Machine Learning Approach to Determine Abrasive Belt Wear of Wide Belt Sanders Maximilian Bundscherery Thomas H. Schmitty Sebastian Bayerly Thomas Auerbachzand Tobias Bocklety.pdf

共4页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

An Acoustical Machine Learning Approach to Determine Abrasive Belt Wear of Wide Belt Sanders Maximilian Bundscherery Thomas H. Schmitty Sebastian Bayerly Thomas Auerbachzand Tobias Bocklety

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: