JP2023552680A - Apparatus and method for continuously detecting yarn defects in a spinning machine - Google Patents

Abstract

A device for detecting defects in drawn roving (R) being processed in a spinning machine includes an acquisition means (60), digitally processes the acquired image (I), and uses a machine learning detection algorithm to detect defects. processing means (70) for detecting defects.

Description

The present invention relates to the field of textile fiber processing, and in particular to a machine and method for detecting defects in the product of a spinning preparation process. In particular, the subject matter of the invention is a method and a device for detecting tangles or knots, commonly referred to as "neps", in yarn produced on a ring spinning machine.

As is known, a spinning machine can process a bobbin of roving to obtain a spool of yarn after drawing and twisting the roving.

To achieve this purpose, the spinning machine is composed of a frame, a drawing device and a rail. The frame extends along a longitudinal axis and supports a creel from which a bobbin of roving is suspended. The drawing device is supported by a frame and consists of a set of connected longitudinally extending cylinders through which the roving is drawn. The rail carries along its longitudinal axis a row of spindles which rotate about their vertical axes and from which the drawn and twisted yarn is wound.

It is well known that yarn defects have an undesirable effect on the appearance of textiles, especially in the case of colored textiles. Therefore, there is a strong need in the industry to monitor the extent and frequency of defects, particularly the extent and frequency of fiber entanglements.

To date, there are two industrial methods for detecting defects from yarn.

In the first method, a sample consisting of several spools of thread taken from a spinning machine is analyzed in a laboratory, typically using capacitive sensors that can detect changes in mass along the thread. This is done by defining the type and frequency of entanglements. Although this method is generally reliable, since it is carried out after the end of yarn production, it is not possible to intervene in processing parameters to improve yarn quality or to uncover the causes of defects. For example, USTER® Tester 5-S800, manufactured and sold by the company Uster Technologies AG, is often used.

In the second method, a detection module is used, which is based on a capacitive or optical sensor placed on the spindle for winding the thread, and a blade capable of physically removing the thread entanglements. Equipped with Even in this case, it is not possible to trace the cause of the defect and intervene according to the cause.

Other methods apply to the yarn winding stage, a process that takes place downstream of the spinning machine. For example, the Uster® Quantum 3 tester, based on capacitive sensors, manufactured and sold by Uster Technologies AG, and the YarnMaster Prisma tester, based on optical sensors, manufactured and sold by Loepfe Brothers, are widely used. .

Furthermore, a first objective of the present invention is to detect defects in the yarn in a spinning machine and to change the processing parameters or maintenance procedures of the spinning machine or other upstream machinery to obtain a higher quality yarn. .

The methods described above are not suitable for this purpose as they use capacitive or optical sensors that require regular yarn feeding.

However, between the drawing device and the winding device of the spinning machine, the yarn is subjected to strong vibrations due to the winding and twisting operations that occur downstream.

Some studies acquire images via a video camera and then process the images to detect tangles. For example, several methods are described in the following papers:
- Li Z, Pan R and Gao W. , “Generation of digital yarn blackboard using a series of images”, Textile Research Journal, 2016, 86: 593-603
- Eldessouki M, Ibrahim S and Militky J. A, “Method for measuring thread diameter and its variation based on dynamic and robust image processing”, Textile Research Journal, 2014, 84: 1948-60
- Ling C, Lianying Z, Li C and Xuanli Z. , “Digital image processing of cotton yarn seriplane”, 2010, International Conference on Computer and Information Application, 2010, p. 274-7
- Li Z, Xiong N, Wang J, Pan R, Gao W and Zhang N, "Intelligent processing method to automatically mosaic continuous slub yarn images based on image processing", Textile Research Journal, 2018, 88:2854-66
- Carvalho V, Soares F and Vasconcelos R, "Artificial Intelligence and Image Processing-Based Techniques: Tools for Yarn Parameterization and Textile Prediction", 2009 IEEE Conference on Emerging Technologies & Factory Au tomation, 2009, p. 1-4

However, these methods require periodic feeding of pre-tensioned yarn and are therefore not suitable for the intended industrial purpose.

Finally, some solutions are proposed, for example, China Patent Application No. 111235709, China Patent Application No. 109389583, China Patent Application No. 105386174, German Patent Application No. 102018111648. It is explained in the specification, WO 2019/130209, JP 2018-178282, and DE 102016121662.

Furthermore, the currently known solutions are also unsatisfactory in terms of the reliability of the results. In fact, slight irregularities in the fibers, such as slight enlargement (allargamento) of the fibers in certain areas, are often mistaken as entanglements or other defects.

It is an object of the present invention to provide a method and a device for detecting defects in processed yarns in a spinning machine, which meets the requirements of the industry and overcomes the drawbacks mentioned above with respect to the prior art.

This object is achieved by a method according to claim 1 and a device according to claim 17. The claims dependent thereon specify further advantageous embodiments of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS The features and advantages of the method and device according to the invention will become apparent from the following description, given by way of non-limiting example and in accordance with the figures of the accompanying drawings.

A diagram showing a spinning machine equipped with a detection device according to the present invention Diagram of the spinning machine in Figure 1 Original NEP positive image shown with corresponding pixel profile Positive image of the original NEP shown with corresponding pixel distribution Synthetic NEP positive image

For clarity of explanation, methods and apparatus for detecting nep are described below. However, it will be appreciated that the invention has general application to defect detection as well.

With reference to the accompanying drawings, reference numeral 1 collectively designates the spinning machines of a spinning line, the spinning machines having an extension along a longitudinal axis It is intended to obtain.

The spinning machine 1 includes a frame 2 for supporting elements made up of one or more parts arranged side by side, and a creel 4 supported by the frame 2.

The creel 4 comprises a vertical post 6 and a plurality of longitudinal cross members 8 supported by the post 6 at a predetermined height. The cross member 8 is for supporting a plurality of suspended bobbins of roving B.

Below the crosspiece 8 , ie downstream of the bobbin B, the spinning machine 1 comprises a drawing device 10 supported by a frame 2 .

The stretching device 10 includes a plurality of lower stretching cylinders 12a-12d, typically three or four cylinders. The cylinders are electrically powered, longitudinally extending, made in one piece or in multiple pieces, and made as structurally distinct, parallel-aligned segments.

The stretching device 10 further includes a plurality of pressure arms 14 arranged in parallel in the longitudinal direction. Each pressure arm carries an upper idle pressure roller 16.

The pressure roller 16 is connected to the pressure cylinders 12a to 12d to form drawing pairs, and the roving passes through the drawing pairs so that the peripheral speed of each drawing pair increases from upstream to downstream. It is stretched by increasing the direction.

The spinning machine 1 further comprises a winding device 17 arranged below the drawing device 10 and immediately downstream of the first drawing cylinder 12a.

The winding device 17 includes a thread guide assembly 18, and the thread guide assembly 18 includes a support 20 connected to the frame 2 and a plurality of thread guides 22 supported by the support 20 and arranged in parallel in the longitudinal direction. Be prepared.

The winding device 17 further comprises a rail 23 located below the thread guide assembly 18 , ie downstream of the thread guide 22 . The rail 23 is supported by the frame 2 and is movable up and down by reciprocating motion.

The winding device 17 further comprises a plurality of spindles 24 arranged longitudinally side by side along the rail 23, each spindle 24 being rotatable about a respective vertical axis.

During normal use of the spinning machine 1, the roving wound on a given bobbin B enters the drawing device 10 through a first path segment and is withdrawn therefrom. The drawn roving R passes through a second path segment S2 between the drawing device and the corresponding spindle 24 and through the corresponding thread guide 22. The yarn resulting from drawing and twisting the roving is wound onto a tube attached to a spindle to form a spool of yarn.

According to the invention, the detection area 50 is defined as the area intersected by the second path segment S2 of the drawn roving R between the drawing device 10 and the winding device 17.

The drawn roving passing through the detection region 50 is subjected to processing for defect detection, and in particular for the detection of tangles and knots, commonly referred to as "neps."

The detection process includes an acquisition step of acquiring an image of at least one segment R* of the roving R that is moving in the detection area 50.

For this purpose, the spinning machine comprises acquisition means 60 adapted to acquire an image I of at least one segment R* of the roving R while moving through the detection area 50, the acquisition means e.g. Equipped with digital and video cameras.

Furthermore, the spinning machine 1 preferably comprises illumination means 62, which are adapted to illuminate at least one zone of the detection area 50 comprising the segment R* of the roving R. The illumination means 62 provide illumination, for example by means of warm or cold light or, in a variant, by means of infrared radiation.

The detection process further includes the step of processing the image I acquired by the acquisition means 60, performed by the processing means 70.

Said processing means 70 are arranged to detect NEPs by means of an adaptive or machine learning type detection algorithm, where the detection algorithm is preferably based on the Viola-Jones method, in particular an algorithm of the Haar-cascade type. It's an algorithm. An algorithm based on the Viola-Jones method is the ``Rapid object recognition method using simple enhanced cascades'' presented by Paul Viola and Michael Jones at the Conference on Computer Vision and Pattern Recognition (2001). In a paper entitled ``Detection'' The teachings in that paper regarding the implementation of the algorithm are expressly incorporated into this disclosure.

The detection algorithm is trained using a set of positive images (Ip), such as images showing that segment R* of drawn roving R has neps, and images showing that segment R* of drawn roving R does not have neps. It is based on a set of negative images (In) such as .

According to a first embodiment (referred to as "with native neps"), the positive images Ip in the set of positive images are the training images in which segments with neps and segments without neps are depicted. It is determined by starting with images of drawn rovings, for example images acquired during normal use of a given spinning machine, and processing the pixel profile of each image. The pixel distribution is obtained by adding a bright pixel to each row of the image and selecting as a positive image an image in which at least one peak of the pixel distribution exceeds a threshold, the threshold being determined by e.g. the standard deviation and the mean value. determined (Figures 3a and 3b).

Additionally, positive images are preferably further selected by excluding images where a peak is caused by a single frayed fiber. The further selection is done by an additional selection algorithm or manually.

According to a further embodiment (referred to as "with synthetic nep"), the positive image Ip is constructed digitally and is formed by a semicircular or semielliptical image. The positive image Ip is preferably formed by an image of a vertical semicircle or a vertical semiellipse (FIG. 4), for example, by an image of a vertical semiellipse whose long axis and short axis are different in length. Ru.

According to yet another embodiment (referred to as "hybrid"), the positive image Ip comprises a positive image with original nep and a positive image with synthetic nep, i.e. by combining the two previously described embodiments. There is.

According to a further embodiment of the invention, the processing means 70 comprises an adaptive or machine learning system of the Convolutional Neural Networks (CNN) type and/or the Recurrent Neural Networks (RNN) type. is configured to detect NEPs using a detection algorithm.

During normal use of the spinning machine, when the roving R is moving through the detection area 50, the acquisition means 60 continuously acquires images I of segments R* of the roving R.

The image I is processed by processing means 70 to detect the frequency of occurrence of neps, preferably also the shape of the neps, and to classify the neps by type. The detection is carried out by a detection algorithm of the machine learning type, particularly preferably of the Haar-cascade type based on the Viola-Jones method or of the type Convolutional Neural Networks (CNN) and/or of the Recurrent Neural Network (RNN) type. done by method.

Based on these findings, as part of a method for managing a spinning machine or a spinning line comprising a spinning machine and its upstream equipment, processing parameters of the spinning machine may be adjusted for the quality of the roving or Due to the processing performed by the equipment upstream of the spinning machine, changes are made or maintenance work is performed on the spinning machine or on the equipment upstream of the spinning machine. Equipment upstream of the spinning machine is, for example, a card, a combing machine, a drawing frame, and a roving frame. The processing parameters of the spinning machine are, for example, the twist of the yarn, the drawing and/or predrawing of the roving, the type and weight of the rings, the hardness of the shaft rubber, the cylinder gauge, the pressure on the shaft, and the production rate. Also, the processing parameters of the equipment upstream of the spinning machine, such as twisting, drawing and/or pre-stretching, hardness of the shaft rubber, cylinder gauge, pressure on the shaft, production rate in the roving frame, production speed in the drawing frame, etc. Production rate and draw assembly gauge, percentage of rejects, number of strokes and draw unit gauge in the combining machine, production rate, removal of neps and debris in the card and blow room line. It is.

Innovatively, the apparatus and method for detecting defects in a spinning machine according to the present invention meets the needs of the industry and overcomes the drawbacks mentioned above.

In fact, the detection is carried out continuously on the roving being processed in the spinning machine, so that interventions can be made to change the processing parameters of the spinning machine or upstream machinery, or maintenance can be carried out to improve the quality of the yarn. can intervene.

Advantageously, the invention also makes it possible to distinguish between major defects and other minor defects to an industrially acceptable degree, thus ensuring good reliability of the results.

Furthermore, the tests carried out showed a good agreement between the measurements obtained by the device of the invention and tests based on testers currently in common use, as mentioned in the introduction. .

Furthermore, the image processing according to the invention is advantageously very fast, allowing continuous detection and rapid treatment of defects to improve production.

It is obvious that a person skilled in the art can make modifications to the above-described method and apparatus to meet unspecified needs, and all such modifications fall within the scope of protection defined in the claims.

Claims (17)

A method for detecting defects in drawn roving (R) being processed in a spinning machine, the method comprising:
acquiring an image (I) of the drawn roving (R) moving within the segment (Sr) of the path (S2);
digitally processing the image (I) to detect the defect;
including;
The image processing step (I) includes detecting defects using a machine learning type detection algorithm,
The training of the detection algorithm is based on a set of positive images (Ip) with defects in the drawn roving segments and a set of negative images (In) with no defects in the drawn roving segments. ,
Method. The processing step includes:
detecting the frequency of occurrence of the defects along the roving (R);
The method according to claim 1. The processing step includes:
detecting the type of defect;
The method according to claim 1 or 2. Starting from training images of a drawn roving with defective and defect-free segments, a positive image (Ip) is determined by processing the pixel distribution of each image;
said pixel distribution is obtained by adding bright pixels to each row of the image and selecting as a positive image an image in which at least one peak of the pixel distribution exceeds a threshold;
A method according to any one of claims 1 to 3. The threshold value is determined by the standard deviation and the mean value of the distribution,
The method according to claim 4. The training roving image is obtained during normal use of a predetermined spinning machine;
The method according to claim 4 or 5. Positive images are further selected by excluding images where the peak is caused by a single frayed fiber.
7. A method according to any one of claims 4 to 6. said further selecting is performed by a further selection algorithm;
The method according to claim 7. said further selecting is done manually;
The method according to claim 7. The positive image (Ip) is digitally generated and is formed by a vertical semicircle or a vertical semiellipse, for example, by a vertical semiellipse whose long axis and short axis are different in length. be done,
A method according to any one of claims 1 to 9. A portion of the positive image (Ip) is obtained by a machine learning-based detection algorithm,
The remaining part of the positive image (Ip) is obtained by digital construction.
A method according to any one of claims 1 to 9. The detection algorithm is a Haar-cascade type detection algorithm,
12. A method according to any one of claims 1 to 11. The Haar-cascade type detection algorithm is based on the Viola-Jones method,
13. The method according to claim 12. The algorithm is a Convolutional Neural Network (CNN) type algorithm and/or a Recurrent Neural Network (RNN) type algorithm.
12. A method according to any one of claims 1 to 11. A method for managing a spinning machine (1) in a spinning line, the method comprising:
A method for detecting defects in drawn roving (R) according to any one of claims 1 to 14,
a subsequent step of changing processing parameters of the spinning machine or performing maintenance work on the spinning machine;
including,
Method. A method for managing a spinning line including a spinning machine (1) and equipment for processing fibers upstream of the spinning machine, the method comprising:
A method for detecting defects in the spinning machine (1) according to any one of claims 1 to 14;
a subsequent step of changing a processing parameter of at least one said piece of equipment upstream of said spinning machine or performing maintenance work on at least one said piece of equipment upstream of said spinning machine;
Method. A device for detecting defects in drawn roving (R) being processed in a spinning machine,
Acquisition means (60 ), wherein the detection area (50) is traversed by a segment (Sr) of the path (S2) of the roving (R);
processing means (70) operatively connected to the acquisition means for digitally processing said image (I) to detect said defects;
including;
The processing means (70) are configured to detect defects using a machine learning type detection algorithm, the learning of the detection algorithm being based on a positive image (Ip) in which the drawn roving segment is defective. and a set of negative images (In) in which the depicted drawn roving segments are free of defects.
Device.

Images