WO2022063373A1

WO2022063373A1 - Optical assessment of packaging film welds

Info

Publication number: WO2022063373A1
Application number: PCT/DK2021/050292
Authority: WO
Inventors: Jan Hansen
Original assignee: Qubiqa A/S
Priority date: 2020-09-23
Filing date: 2021-09-22
Publication date: 2022-03-31

Abstract

Packaging packages in plastic film is an important part of industrial and commercial processes. Packaging leakage is expensive, but improved package welding is difficult and time-consuming and so also expensive. This is solved by providing a packaging apparatus (100) with plastic film welding means, further having: an optical sensor (110) to provide images (164) of a film weld (11, 12) produced by the packaging apparatus (100) and transmit said images as an image signal, a database (160) having contouring instructions (161 ) and a tensile algorithm (165), a processor (121 ) for receiving said image signal and adapted to: - generate weld data (170) from the image of said image signal by using said contouring instructions (161 ), where contour lines (171 ) of said weld data are generated for locations in the image with a high contrast, and where at least some of said contrasts correspond to boundaries between areas of welded film and unwelded film, - predict tensile strength of a weld (11, 12) based on said contour data (170) using said tensile algorithm (165), and - transmit a fault signal if the predicted tensile strength of a weld is below a certain tensile strength threshold.

Description

OPTICAL ASSESSMENT OF PACKAGING FILM WELDS

FIELD OF THE INVENTION

Packaging apparatus with a fault identification module and a method of using the fault identification module when packaging with the packaging apparatus.

BACKGROUND OF THE INVENTION

Industry relies heavily on distributed production of goods which includes efficient logistics of goods, such as between a production site and a consumption or use site. It is important, therefore, to be able to package these goods efficiently for storage and transport. Many goods are packaged by wrapping with plastic film. One useful technique is wrapping a pre-compressed item in plastic film. The payload expands in the wrapping and thereby produces a rigid overall package. Another common technique is shrink-wrapping.

Welding plastic film is important to create the closure of the package and may also provide the correct final tension on the plastic film. Sufficiently strong welds may initially seem to be achieved using different combinations of welding settings, that is, welding time, welding temperature, and so on.

The technician has to periodically review whether the films seem to be delaminating or tearing around the weld before the package is shipped.

However, for packaging machines, defects in the welds may show up outside the production facilities, and thus it may be very difficult to use such defects to pinpoint the problematic setting. As an example, a seemingly adequately wrapped package may be sent away only to burst open in a truck because of an insufficiently fused weld, spreading the payload throughout a truck. Since packaging may have branding on it, tearing or bursting at a retail site may be detrimental to the brand owner. To remedy packages bursting or tearing, the quality of all welds may be increased, such as by increasing welding time. This reduces packaging efficiency however, by wasting time over-welding packages while not actually ensuring better quality.

There is thus a need for improved packaging using plastic film. SUMMARY OF THE INVENTION

In an aspect of the invention, there is provided a packaging apparatus with plasticfilm welding means, further having:

- an optical sensor to provide an image of a film weld produced by the packaging apparatus and transmit said image as an image signal,

- a database having contouring instructions and a tensile algorithm,

- a processor communicatively coupled with the database, the processor adapted to receive said image signal and adapted to:

- generate weld data from the image of said image signal by using said contouring instructions, where contour lines of said weld data are generated for locations in the image with a high contrast, and where at least some of said contrasts correspond to boundaries between areas of welded film and unwelded film,

- predict tensile strength of a weld based on said contour data using said tensile algorithm, and

- transmit a fault signal if the predicted tensile strength of a weld is below a certain tensile strength threshold.

Thereby, the tensile strength of a weld of a film wrapping can be predicted for every weld and package passing through a packaging line.

This is an improvement over a technician review of the welds, firstly since the technician may not be able to get a proper angle of viewing the welds while the welds are on the packaging line. Such a viewing angle may be best if it is over the welding means or at least very close hereto. Secondly, a technician may not be able to keep up with the packaging speed of the packaging line, in which case there is no improvement achieved at all, and longer welding may be just as beneficial. Thirdly, the technician will experience fatigue and miss important weld quality indicators after a while. Fourthly, assessing weld quality is an experience-based process that takes time to learn.

The benefits achieved are plentiful. Lower costs are achieved since quality is assessed continually, and welding settings can be optimised. The technician can tune the welding parameters while gaining feedback on how the tensile strength is affected in real time. He may then reduce heat or time to improve economic metrics as far as is useful and permissible.

It also allows a higher quality assurance, resulting in fewer packaging defects downstream. Production faults are a problem, but shipping a damaged product to a customer has much higher costs - the defective package takes space in shipping from a sellable package, the fault may produce waste and clean up, return shipping or disposal of the damaged product, and brand damage if the packaging is branded.

The invention works in principle by using image analysis to find topological features of the packaging and/or thermal discolouration areas. Finding the topological features and thermal discolouration is performed using contouring methods, which identifies areas in an image with high contrast, then draws lines along these high contrast boundaries.

The inventor realised that although a weld is not an object boundary per se, the welded film differs from neighbouring unwelded film. For example, the weld is thick and smooth, where boundary region between the weld and non-welded film may be heat-affected and thus affected for having been proximate to the heating process without being compressed.

Further, the boundary region has two adjacent non-joined plastic films whose combined thickness is larger than the thickness of the weld, since the weld is compressed as part of the welding process. Therefore, the boundary films spread away from one another, creating a topological feature.

Furthermore, the opacity of the welded area may be changed by melting.

By plastic film is meant generally polymeric material of a pliable thickness useful for wrapping payloads for shipping, storage and handling, also known as polymeric foils.

In an embodiment, the weld data further comprise weld parameters relating to weld line length data and/or weld area data. Thereby, the tensile algorithm can provide precise predictions of tensile strength.

In a preferable embodiment, the weld data comprising weld line length data and weld area data are combined with the tensile algorithm comprising a trained component. Thereby, the fuzzy information present in weld images resulting from imaging plastic film welds can be even better used and understood by the tensile algorithm.

In an embodiment, the weld data comprise at least one weld parameters from: |weld line length of the longest continuous weld, |summed length of a plurality of weld lines, (length of longest weld line periphery, (summed length of all weld line peripheries, |area of the largest continuous weld area, (summed area of a plurality of weld areas, and (summed horizontal weld line length,

Weld lines are continuous contours, weld peripheries are continuous contours that perform at least a 150-degree bend, preferably at least a 180-degree bend, weld areas are areas fenced at least substantially by contours, and summed horizontal weld line length is the summed contour length along the weld direction.

Other weld parameters can also be contemplated. Secondary weld parameters comprise (8) image brightness mean value, and (9) image brightness standard deviation. These secondary weld parameters are probably not suitable to be the only weld parameters considered, and thus are not presented in the above list. They are very useful as extra parameters however.

In a preferable embodiment, weld data comprise all of parameters 1 , 2, 3, 5 and 6.

In a yet more preferable embodiment, this embodiment is combined with the tensile algorithm comprising a trained component. Thereby, the fuzzy information present in weld images resulting from imaging plastic film welds can be even better used and understood by the tensile algorithm.

In an embodiment, weld data are corrected or informed by one of image brightness mean value, and image brightness standard deviation. Thereby, the invention takes into account differing conditions such as different lighting conditions and different package angles on the outfeed which may produce a different light reflection and thus a different imaging light. In an embodiment, the tensile algorithm takes image brightness mean value and/or image brightness standard deviation into account. In an embodiment, generating weld data from the image comprises a step of separating a weld shape from artefact shapes in the image, and not including artefact shapes in weld data, where said separating is performed by at least one of: comparing the image with at least one known weld to identify relevant image area; and comparing the image with at least one known artefact to correct or ignore relevant image area. Thereby, the optical sensor can transmit a relatively raw image to the computing device, and the computing device can take care of cropping to the useful image section. This furthermore makes the fault detection more robust since errors may be present within the weld region that can be ignored, and the packages may differ slightly in orientation from package to package. Furthermore, this allows making robust predictions regardless of branding or other information on the package, such as the payload being visible beneath the weld.

In an embodiment, a technician interface allows local modifications to adapt at least one of: image contrast thresholds that determine which sensitivities contours are generated for an image to take local lighting and imaging conditions into account; and the tensile algorithm to adapt the mapping between weld data and predicted weld tensile strength, to take into account local welding conditions.

Thereby, the apparatus and method can be adapted to local conditions and the apparatus and method become much more precise in predicting tensile strength.

In an embodiment, the tensile algorithm comprises a trained component, which trained component has been created with a supervised learning method based on a training set of welds, each weld in the training set mapped to a measured tensile strength.

Thereby, very accurate tensile strength prediction has been found to be achievable.

In a preferred embodiment, the tensile algorithm comprising a trained component is combined with the weld data comprising two or more weld parameters.

The trained component may be trained in a variety of artificial intelligence methods, such as using a neural network, machine learning, deep learning, a deep neural network or an artificial neural network. Many of these terms relate to overlapping computational techniques as well. In the following, the invention will be described as using a deep neural network (DNN), which may be a particularly useful technique for the invention.

The DNN is provided with at least one, preferably at least two weld parameters as input. Even more preferably, all weld length and area parameters are provided as input for the DNN.

A series of tests are performed to produce a training set. The tests comprise taking an image of a weld, then testing its actual tensile strength. These tests serve as the supervision output.

The DNN is then tasked with producing a tensile strength prediction based on the weld parameters, and the DNN is then trained using the training set and its weld parameters. The inventor found that surprisingly, very accurate prediction was possible based on weld parameters even though images seem to provide fuzzy data.

By providing a deep neural network, several layers of virtual neurons are provided. This allows the apparatus and method to use the weld parameters to contextualise each other. For example, an image may produce a contour line only on one side of a weld (the top or bottom). The influence of a single weld contour may depend on image brightness standard deviation - if an image is generally bright and so has a low brightness standard deviation, this predicts a low image contrast. A low image contrast may in turn signify an inability to produce good contours. Taking this into account the single-side contour weld in a low-contrast image is perhaps considered favourably by the trained component compared to a single-side contour weld in a high-contrast image, where this may signify a weak weld.

Specific areas of welds can be added as outputs for the training of the component. This can be understood as performing tensile strength tests on sections of weld instead of testing only whole welds. A specifically poor section of weld can be tested and the tensile strength added as a part of the training material for the component. Such a specific area of a weld can be added to train the component to recognise faulty or strong patterns in the weld. This will sometimes in the specification be termed as a strong weld indication or a weak weld indication. In an embodiment of the invention using a trained component, the welded packages are provided with a unique signature that is tied to a weld image and/or weld data. Images and image data are also stored over time for a time, and when a fault is reported for a package and is identified, the database is instructed to consider the associated welded data as a fault weld. This fault weld is then used to retrain the trained component, thus improving accuracy of predicted tensile strength over time.

In an embodiment, the apparatus stores predicted tensile strength of imaged welds in the database, and where the processor is further adapted to analyse the predicted tensile strength of consecutive welds over time to provide a second fault signal if the predicted tensile strength over time fulfils a certain pattern. Thereby, machine wear can be identified before failures begin. Welding jaw fatigue is one type of machine wear that is useful to identify. Simpler and different creeping errors may of course also be pinpointed.

In an embodiment, the apparatus further comprises at least one light source for providing illumination of a weld during providing an image of said weld. Thereby, image quality is ensured. In an embodiment, the weld region is brightly illuminated for imaging. In an embodiment, light sources are located relative to the optical sensor in a manner to produce reflections that are used to improve image contrasts for contour detection.

In an embodiment, the weld data or the tensile algorithm informs at least one welding jaw operational parameter. In an embodiment, the weld data or the tensile algorithm results in the apparatus and method proposing to a technician through an interface at least one welding jaw operational parameter correction. Both of these embodiments allow, by reviewing the continual output quality of the packaging apparatus, the welding process to be improved. A weld with certain characteristics may be indications of specific undesirable welding conditions that can be difficult for the technician to realise. A border region next to a weld may show heat damage, or tearing from heat damage. A part of the weld region may be less transparent than desired, evidence of insufficient welding temperature, time, or pressure. This may also be due to plastic film characteristics, whereby switching to a new film may benefit from changing the welding slightly.

In an aspect, the invention relates to a method comprising: - packaging a payload in plastic film using a packaging apparatus with plastic film welding means,

- providing an image of a weld produced by the packaging apparatus using an optical sensor,

- transmitting said image to a processor as an image signal, the processor being communicatively coupled with a database having contouring instructions and a tensile algorithm,

- generating, by the processor, weld data from the image of said image signal by using said contouring instructions, where contour lines of said weld data are generated for locations in the image with a high contrast, and where at least some of said contrasts correspond to boundaries between areas of welded film and unwelded film,

- predicting tensile strength of a weld based on said contour data using said tensile algorithm, and

- transmitting a fault signal if the predicted tensile strength of a weld is below a certain strength threshold.

In an embodiment, the film used for packaging payloads in the packaging apparatus comprises an outermost film layer being at least substantially transparent and another film layer being at least substantially opaque.

Thereby, the invention can review weld quality by imaging both layers while ignoring artefacts behind the weld.

BRIEF DESCRIPTION OF THE DRAWINGS

In the following, example embodiments are described according to the invention, where:

Fig. 1 is a schematic of a packaging apparatus of an embodiment of the invention,

Fig. 2 is a schematic of imaging a weld of an embodiment of the invention,

Fig. 3 is a schematic of a computing device of an embodiment of the invention,

Fig. 4 is an image of a plastic film weld with other artefacts present imaged of an embodiment of the invention, and Figs. 5A-5I show different weld images with their matching generated contour lines of an embodiment of the invention.

DETAILED DESCRIPTION

In the following, the invention is described in detail through embodiments hereof that should not be thought of as limiting to the scope of the invention.

Fig. 1 is a schematic of a packaging apparatus 100 according to an embodiment of the invention.

The packaging apparatus has an optical sensor 110 for imaging film welds, a computing device 120 to use images for analysis, jaws 130, 131 with cutting means 132 and welding means 133, 134 to weld and cut film 10, and film rolls 140, 141 to hold the film ready for packaging. The packaging apparatus is placed between an infeed line 20 and an outfeed line 30.

The packaging apparatus 100 operates as follows. The packaging apparatus 100 receives a payload 1 from an infeed line 20. The payload may be treated in some manner to prepare it for packaging. For example, it may be compressed a certain degree. This both reduces size and so freight costs, and provides an outwards push when packaged that fills up the package 2 to provide rigidity.

The payload 1 is then pushed against the plastic film 10, which is maintained tight along its path, however, it is easily pulled along by the payload 1 movement. A number of rollers 142, 143, 144, 145 facilitate efficient packaging of the payload 1. The payload 1 then pulls the plastic film 10 along with it onto the outfeed line 30.

When the payload has passed the jaw by a predetermined amount (not shown), the jaws 140, 141 are brought together behind the payload along movements A and A’ respectively. The plastic film 10 is retained firmly between the jaws 130, 131 , and then welded along two lines, one for each welding means 133, 134. The welding means may be ultrasonic or thermal. While the jaws 130, 131 hold the plastic film, the cutting means 132 cuts between the two weld lines.

The cutting means shown comprises matching knife and anvil, where the knife is moved along the direction B to cut the film. The dual welding and cutting jaw structure allow producing an end-weld 12 for an outgoing package 2 while also welding a front-weld 11 in preparation for the next payload 1 in one single operation and is thus an efficient packaging apparatus structure. However, other welding and cutting arrangements are contemplated within the scope of the invention.

When the jaws 130, 131 have welded and cut the plastic film as instructed, the package 2 leaves by an outfeed line 30. It is now typically ready for the next step, such as shipping or further packaging.

The apparatus further has an optical sensor 110 located between the jaws 130, 131 and the outfeed line 30. As long as it is located near at least one of these, it may perform its function. The optical sensor 110 is adapted to film or capture images of the plastic welds. It can capture both front-welds 11 and end-welds 12. In the shown embodiment, it can capture both welds at once and using only the one optical sensor. In an embodiment, two or more optical sensors may be used. In an embodiment, an imaging angle may be used that can capture both front-welds and end-welds during operation of the packaging apparatus.

In any case, an image is provided by the optical sensor 110 of a weld 11 , 12, and transmitted as an image signal.

A computing device 120 receives the image signal for analysing. Depending on what the computing device 120 finds in the image, a fault signal is be transmitted, such as to a technician that may then correct operation of the packaging apparatus 100.

Different faults can be inferred from different types of analyses. For example, an uneven weld can be a sign of jaw fatigue - if the weld is strong along the sides with weak spots emerging in the middle over a series of welds, this indicates that the jaws are getting fatigued.

Overall, weld density, shape and stability over time may inform of a variety of different faults. Examples of welds will be discussed in relation to Figs. 4-5.

Fig. 2 is a schematic of imaging a weld 12 of a package 2 on an outfeed line 30. An optical sensor 110 provides an image of the weld region 111 being the area of the package having the weld. Film overlap lines 13, 14 indicate where the bottom film and top film ends, respectively. The top film is outermost and has a skirt below the weld 12. The weld 12 is shown with interrupted lines 12A, 12B that are also bent and crooked.

The faults in the weld 12 are exaggerated for illustrative purposes.

To improve imaging, a plurality of light sources 150, 151 , 152, 153 are provided. Having at least one light source adapted to provide adequate light for the imaging helps both illuminating the weld sufficiently that the weld gains fidelity, as well as developing the appropriate contrasts to identify the welds.

In an embodiment of the invention having a plurality of light sources 150, 151 , 152, 153, the apparatus comprises a frame 155 that holds the plurality of light sources at a mutually predetermined distance and may thus help ensure proper lighting for correct imaging. In an embodiment having at least one light source 150, the apparatus has a frame 155 holding the light source and optical sensor 150 at a predetermined distance.

The package may have artefacts that the apparatus has to take into account. For example, the package may have a communication artefact 3 such as branding or product information that overlaps the welded area. This may make analysis more difficult. Another type of artefact is error artefact 4. Such error artefact 4 may be a tear or dirt or even a blemish on a communication artefact. Such error artefact is firstly ignored when identifying welds and tensile strength of the weld, and may secondly produce fault signals of their own type.

Fig. 3 is a schematic of the computing device 120 of the invention. The computing device has a processor 121 , a communications element 122 and a database 160.

The database stores contour instructions 161 which, when executed by the processor, identifies contours in an image provided by the optical sensor. Contours are generated based on image contrast, and indicates generally an object boundary.

The contour lines are then identified by using the contour instructions and weld data is produced. The weld data 170 is also stored in the database, at least while the analysis of the specific weld is performed.

Weld data 170 comprise firstly the contour lines 171 produced by the processor executing the contour instructions 161. Preferably, producing the weld data 170 comprises a step of object recognition. The object recognition allows identifying different objects in the image 164 to help create accurate weld data 170.

Firstly, producing weld data 170 may have a step of mapping a weld target object

162 to the image 164, the weld target being for example an elongate pattern indicating the weld in the relevant area of the image 164. Contour lines may then be limited to this identified relevant area.

Secondly, producing weld data 170 may have a step of identifying one or more image artefacts 163 and subtracting the artefact 163 from the weld image 164 or otherwise modifying the image based on the artefact. For example, if a large branding logo or other feature is present, the database may store this image artefact

163 and map it to the image 164 of the package, then subtract it therefrom to allow better comparing welds among areas where the background colours are different. If a light-related artefact such as shine from a light source produces a sharp contrast, this may be identified and ignored.

After the contour lines 171 are created for a weld, further weld data 170 may be created for the weld based on this and other data.

In a preferred embodiment, the produced contour lines 171 of a weld are analysed to identify at least one contour parameter 172. Contour parameters comprise weld contour lengths, weld area and image parameters, relating to the circumference of an identified weld, the area of an identified weld, and circumstances relating to the image of an identified weld, respectively. Depending on the imaging conditions, it may not be effective to use only a single parameter to adequately characterise the weld.

Weld contour length relates to a sum of a plurality of individual weld contour lengths and/or a selection among these, such as the length of the longest weld contour length.

Weld area relates to a sum of a plurality of areas at least substantially fenced by contour lines, and/or a selection among these, such as the size of the largest area at least substantially fenced by contour lines. In an especially preferable embodiment, the weld data comprises both the mentioned contour parameters.

Image parameters help correct the image and take different conditions into account, and comprise data such as image brightness mean value and image brightness standard deviation.

The database further comprises a tensile algorithm 165. Based on the weld data, a predicted tensile strength 166 is calculated by a tensile algorithm 165.

In one embodiment, the tensile algorithm 165 first sums the contour lengths to find a combined length of all contours deemed to be relevant to the weld for a given image. Then, based on the combined length, an estimated tensile strength for the imaged weld is calculated based on a function of tensile strength per contour distance.

The function mapping tensile strength to contour distance may be experimentally determined using historical test samples.

In other embodiments, a plurality of contour parameters are used as input variables to calculate a predicted tensile strength. These can either be a plurality of variables of a function, or, certain input variables may change the algorithmic interpretation of other input. These more complicated relationships between input variables may play out in a trained component of the tensile algorithm, such as a machine learning component.

For welds whose tensile strength are predicted to be below a certain threshold, a fault signal is transmitted through fault channel 123. This may lead to an apparatus interface, a technician device, an auditory signal emitter or some other device adapted to receive the signal. It can also stop or modify apparatus operation.

Fig. 4 is an image of a plastic film weld 412 with other artefacts present. The image shows a solid weld 412, as well as a number of high-contrast artefacts - a crease 463, a film bend 464 and a film edge 465 all have relatively high contrast, and might produce a contour line. What is more is that the crease 463 matches the weld in contrast, and overlaps it. Ignoring this area may be beneficial. However, the shine 463A that is present in the middle of the crease also indicates weld. The image comprises a weld region 411 signifying cut-out matching the size of images of Figs. 5A-5I portions.

Figs 5A-5I show different weld images 164 with their matching contour lines 172. The original images are at the top while the contour lines are provided beneath. It is generally, and clearly seen from these images how fuzzy and difficult it is to compare and analyse the images to quantitatively predict weld tensile strength.

Fig. 5A shows the strong weld, from Fig. 4. Although the weld is very strong and the contrast is high, the contouring instructions have failed to identify the bottom edge of the weld as an edge. This results from a contrast that is, pixel-for-pixel, too low to meet the contouring threshold of the contouring instructions. If left unremedied, this results in a non-existing area and a lower combined contour length than the weld shown in Fig. 4.

This can be solved by the contouring instructions modifying the contouring thresholds until a satisfactory area is produced, since the contour should always produce an elongated area. Other methods can be used as well, such as evaluating the contour width-wise length, where the single contour line will still provide indication of a strong weld, i.e. , a strong weld indication 176.

Fig. 5B shows a spotty and weak weld in the top image and the matching contour lines in the bottom. The combined length of the contours is longer than for the weld shown in Fig. 5A. It should be noted that when defining a weld area, a substantially fenced area is enough. This is easily implemented by allowing gaps of a certain size such as a few pixels, and still define a weld area within, such as weld areas 573A, 573B, 573C. The individual contours are short and the individual areas are small. This provides a weak weld indication 175.

Fig. 5C shows a weld with an interrupted section 581 providing a weak weld indication 175. Furthermore, as is seen in the bottom panel of Fig. 5C, the contouring instructions have failed to identify the bottom line of a weld area 512C. The solution depends on other parameters. It can be solved by modifying the contouring contrasts of the contouring instructions to produce an area. However, it is also possible that the bottom contour lacks because the area around the weld is weakened in some way, which should result in a weld weak indication. In an embodiment, when the contouring instructions modify the contouring thresholds with which a subset of image contours are generated, contours generated with modified thresholds are attributed correspondingly different tensile strength evaluations.

Fig. 5D shows a dashed weld whose overall shape is relatively uninterrupted, providing a strong weld indication.

Fig. 5E shows a dashed weld whose overall shape is more interrupted than the shape of the weld of Fig. 5D. The contouring instructions have failed to identify certain of the weld spots 512E. This weld may be adequately strong.

However, it is at least degrading from right to left. If this pattern is characteristic of a whole weld or at least a transition of a whole weld, it may signify welding jaw fatigue or otherwise a machine fault. It may thus prompt a secondary fault signal.

Fig. 5F shows a continuous bottom weld edge and an interrupted but present top weld edge. Due to the high contrast in the image, several contour lines clearly emerge that are associated. This should result in a strong weld indication even though there are no defined weld areas.

Fig. 5G shows perhaps the strongest weld yet. The combined contour is long, however, no area is defined. The weld of Fig 5G does show what is termed a periphery. A periphery in this sense is a contour that makes a 180-degree bend, which signifies a proper weld.

Figs. 5H and 5I show two images with different brightness. Fig. 5H illustrates a weld image with a higher brightness, a higher mean image brightness and a lower image brightness standard deviation compared to Fig. 5I. Both of these may be used to interpret the weld image better for contour instructions as well as to contextualise weld data for predicted tensile strength calculations.

Claims

|1. A packaging apparatus (100) with plastic film welding means, further having:

- an optical sensor (110) to provide images (164) of a film weld (11 , 12) produced by the packaging apparatus (100) and transmit said images as an image signal,

- a database (160) having contouring instructions (161 ) and a tensile algorithm (165),

- a processor (121 ) communicatively coupled with the database, the processor adapted to receive said image signal and adapted to:

- generate weld data (170) from the image of said image signal by using said contouring instructions (161 ), where contour lines (171 ) of said weld data are generated for locations in the image with a high contrast, and where at least some of said contrasts correspond to boundaries between areas of welded film and unwelded film,

- predict tensile strength of a weld (11 , 12) based on said contour data (170) using said tensile algorithm (165), and

2. The packaging apparatus according to claim 1 , where weld data further comprise weld parameters relating to weld line length data and/or weld area data.

3. The packaging apparatus according to any of claims 1 -2, where weld data are corrected or informed by one of

- image brightness mean value, and

- image brightness standard deviation.

4. The packaging apparatus according to any of claims 1 -3, where generating weld data from the image comprises a step of separating a weld shape from artefact shapes in the image, and not including artefact shapes in weld data, where said separating is performed by at least one of:

- comparing the image with at least one known weld to identify relevant image area, and - comparing the image with at least one known artefact to correct or ignore relevant image area.

5. The packaging apparatus according to any of claims 1 -4, where a technician interface allows local modifications to adapt at least one of:

- image contrast thresholds that determine with which sensitivities contours are generated for an image to take local lighting and imaging conditions into account, and

- the tensile algorithm to adapt the mapping between weld data and predicted weld tensile strength, to take into account local welding conditions.

6. The packaging apparatus according to any of claims 1 -5, where the tensile algorithm comprises a trained component, which trained component has been created with a supervised learning method based on a training set of welds, each weld in the training set mapped to a measured tensile strength.

7. The packaging apparatus according to any of claims 1 -5, where the apparatus stores predicted tensile strength of imaged welds in the database, and where the processor is further adapted to analyse the predicted tensile strength of consecutive welds over time to provide a second fault signal if the predicted tensile strength over time fulfils a certain pattern.

8. The packaging apparatus according to any of claims 1 -7, further comprising at least one light source for providing illumination of a weld during providing an image of said weld.

9. The packaging apparatus according to any of claims 1-7, wherein the weld data or the tensile algorithm informs at least one welding jaw operational parameter.

10. A method comprising:

- packaging a payload in plastic film using a packaging apparatus with plastic film welding means,

- providing an image of a weld produced by the packaging apparatus using an optical sensor, - transmitting said image to a processor as an image signal, the processor being communicatively coupled with a database having contouring instructions and a tensile algorithm,

11 . The method according to claim 10, wherein the film used for packaging payloads in the packaging apparatus comprises an outermost film layer being at least substantially transparent and another film layer being at least substantially opaque.