PESONEN, Jukka (Tekniikantie 14, Espoo, FI-02150, FI)
HAMINA, Sasu (Tekniikantie 14, Espoo, FI-02150, FI)
HEIKKONEN, Jukka (Tekniikantie 14, Espoo, FI-02150, FI)
PESONEN, Jukka (Tekniikantie 14, Espoo, FI-02150, FI)
HAMINA, Sasu (Tekniikantie 14, Espoo, FI-02150, FI)
CLAIMS
1. A method for sorting natural objects into different quality classes (C) based on their properties, which method comprises defining for each object (1) to be sorted a parametre set (s) which comprises a number of parametres that represent the properties of the object, and determining the quality class of the object on the basis of the parametre set, characterized in that the parametre sets (s) of a number of objects (1) are divided into groups (g) of similar parametre sets, at least one group is selected to each quality class (C) , and groups that have not yet been classified are then assigned to the quality classes based on their similarity to the groups that have al- ready been classified, until all groups are classified, so that the quality class of an individual object is determined by the quality class of the group which represents the parametre set of the object, using in the method mathematical distances between the multi-dimensional parametre sets as the measure of similarity between the parametre sets and between the groups .
2. The method in accordance with claim 1, characterized in that the division of the groups (g) into quality classes (C) is changed by moving groups which define the boundaries of the quality classes to adjacent quality classes.
3. The method in accordance with claim 1 or 2, characterized in that a new quality class (C) is added to the classification by selecting a group (g) to the new quality class and assigning groups (g) that are similar to that group to the new quality class .
4. The method in accordance with any one of claims 1 to 3, characterized in that the parametre sets (s) are arranged into a map (4) for as- signing them to groups (g) and further to quality classes (C) , so that the distance between two pararae- tre sets on the map depends on the original mathematical distance of the parametre sets. 5. The method in accordance with any one of claims 1 to 4, characterized in that the groups (g) are divided into quality classes (C) according to the desired share of each quality class in the production. 6. The method in accordance with any one of claims 1 to 5, characterized in that the selection of a group (g) to a quality class (C) is made on the basis of sensory evaluation of a sample object (1) which represents the group. 7. The method in accordance with any one of claims 1 to 6, characterized in that the method comprises sorting of timber, for example boards
(D •
8. The method in accordance with any one of claims 1 to 7, characterized in that the method comprises sorting of stoneware, for example slabs of natural stone (1) . |
SORTING METHOD
FIELD OF THE INVENTION
The invention relates to sorting different kinds of piece goods, such as boards, logs and stone- ware, in an industrial production process. Specifically, the invention relates to sorting the goods into different quality classes according to various properties associated to each good.
BACKGROUND OF THE INVENTION
When handling piece goods in the industry, and specifically objects that originate from the nature, the goods must typically be sorted into different quality classes based on their properties. For ex- ample, boards are sorted at sawmills on the base of e.g. straightness, branching, colour and dimensions.
The division into quality classes can be first of all performed on a sensory basis. In this case, the specific determined individual properties are not used as the quality criteria so much as the general visual impression formed by the evaluator about the object. However, the sensory classification requires a sufficient number of human resources continuously tied to the process, which produces ex- penses. Furthermore, the outcome of the classification inevitably depends at least to some extent on the evaluator, which produces deviation in the final classification. In some embodiments, the sensory evaluation may also be too slow a process. Sensory evaluation can be replaced with measurements. Modern, for example optical, measuring systems are able to quickly gather a great number, even dozens or hundreds, of units of measurement data about the properties of objects passing on the production line without stopping the object. By comparing the
measurement data to a rule base precreated on the base of test measurements, each object can be quickly classified into a specific quality class, and the means for sorting the objects can be controlled accordingly. Measurement and classification of the objects and their sorting according to the classes can be arranged as a computer-controlled automatic process which, besides increasing the speed and accuracy, saves human resources. For each class, the rule base may comprise limit values which are specific for each measured property, and the result must fit within these limits. For example in the case of boards, presence in a specific class may require that the board is not allowed to have more than 3 healthy branches for the length of one metre, and that the entire board does not comprise any branches with a diameter of more than 20mm. Different properties may also be stressed with different coefficients, so that a larger deviation can be allowed for one property, if another and a more signifi- cant one is correspondingly of a specifically good quality. Furthermore, the higher quality classes may comprise some absolute quality criteria.
However, even the rule-based quality classification comprises many problems. First of all, it may be difficult to select the most suitable measurement parametres for each application, and furthermore, their appropriate conditions, from the vast number of measurement data acquired from different measuring techniques, such as cameras or X-ray devices. For ex- ample, in aiming to create a quality class comprising objects which have a good overall impression, as in the case of the sensory classification, it is extremely difficult to find all possible combinations of the measurement data and the respective values which would apply to said class. Creating the rules there-
fore requires an extensive amount of work in the form of test measurements and their analyses.
Secondly, a rule-based system is inflexible when making changes to the classifications. For exam- pie, in changing the conditions of the measurement pa- rametres in a multi-class classification so that objects are moved from one class into another, there may occur at the same time unwanted shifts between the other classes. Creating an entirely new quality class typically requires that all classification rules are reset. However, there are many situations where a quick fine adjustment between classes would be desirable because of the starting material, conditions or market factors. For example, expanding the best qual- ity class slightly so that a higher price could be set for an ever larger share of the products requires heavy measures in order to be able to reset the rules. It should also be noted that the use of the known classification systems requires good knowledge of the numerical data included in the measurement results, and therefore often necessitates a lengthy education and know-how acquired by experience.
OBJECTIVE OF THE INVENTION The objective of the invention is to provide a method for sorting natural piece goods, wherein the classification principles can be created easily and quickly compared to the prior art, and the classification system is also flexible to changes in the classi- fication.
SUMMARY OF THE INVENTION
The invention is characterized by what has been presented in claim 1. The method of the invention for sorting natural objects into different quality classes based on
their properties comprises defining, for each object to be sorted, a parametre set comprising several pa- rametres which represent the properties of the object, and determining the quality class of the object on the base of the parametre set. In this context, the term "natural" signifies that the objects to be sorted are made from material which is acquired from the nature, such as for example wood or stone. These objects typically have a vast variety of properties which affect the quality of the end-product and which may exhibit a relatively large deviation. Determining the classification principles with the traditional methods is in this case extremely challenging. Said parametres can be defined using any measuring method, including for example different optical measurements conducted without contacting the object, and other remote measurements. Some of the parametres can also be defined on a sensory basis. The parametres may represent the appearance of the object and its internal and material properties, for example the dimensions, shape or density of the object. The advantages of the invention compared to the solutions of the prior art become evident when there are several parametres, for example at least three. The maximum number of the parametres that may be included in a parametre set is not limited in the invention; instead, there may be any desired number of them, for example several dozens.
In accordance with the invention, the method comprises dividing the parametre sets of a number of objects into groups of similar parametre sets, selecting at least one group to each of the quality classes, and then assigning groups that have not yet been classified to the quality classes based on their similarity to the groups that have already been classified, until all groups are classified, so that the quality class of an individual object is determined according
to the quality class of the group which represents the parametre set of that object. Further in accordance with the invention, mathematical distances, for example Euclidean distances, between the multi-dimensional parametre sets are used in the method as the measure of similarity between the parametre sets and the groups .
First of all, the method of the invention provides valuable advantages in creating a classifica- tion system. Processing the objects in groups of similar objects makes creating the classification simpler more effective. When at least one group has been, on some basis, assigned to a specific class, the groups that are the most similar to the first group can be assigned to the same class up to the desired quantity, or until another class that has already been formed becomes more suitable. In other words, the entire sorting process merely requires that only one group is selected for each quality class, so that the classifi- cation can then be completed solely according to the similarities between the groups, without taking any standpoint as to the values of the parametres representing individual objects. When the mutual similarities between the parametre sets and the groups are de- fined using mathematical distances, the possibly deviating opinions of individual persons about the similarities will not interfere with the sorting. On the other hand, the method in accordance with the invention provides a classification based on the similarity between the objects, specifically evaluated as a whole, which is quite a challenging objective for example in the sensory evaluation, and becomes all the more difficult as the number of parametres taken into account in the sorting process increases. Furthermore, as each class comprises several groups, they have precise boundaries.
Secondly, a quality classification based on the groups and similarities is extremely flexible when making changes in it. In one preferred embodiment of the invention, the division of the groups into quality classes is changed, if necessary, by moving the groups which define the boundaries of the quality classes to adjacent quality classes. In this manner, assigning the mutually similar groups into the same quality class can be continued as a continuous control, per- formed when necessary, even after the initial classification.
Thirdly, adding new quality classes is easy in the method of the invention. In one preferred embodiment of the invention, a new quality class is added to the classification whenever needed by selecting a group to the new quality class and assigning groups that are similar to the first group to the new quality class. To establish a class, it is therefore sufficient to first select one group that represents the new class, and the class can then be expanded as desired by adding groups that are close to the first group .
In one preferred embodiment of the invention, the parametre sets are arranged for grouping and as- signing to quality classes into a one-, two- or multidimensional map, on which the distance between two parametre sets depends on the original mathematical distance of the parametre sets. In other words, the map is so formed that it retains the original distance re- lations of the parametre sets as accurately as possible. Retaining the distance relations in a completely accurate manner is naturally not possible when moving into a space of smaller dimensions. When the multidimensional parametre sets are visualized in this man- ner as a graphic, for example two-dimensional, map, for instance on a display of a computer, grouping and
quality classification can be performed manually based on the distance relations on the map. Without the two- dimensional manner of presentation, the conceptualization of the multi-dimensional parametre set - there may be for example dozens of individual measured quantities - would be quite difficult. Conversion into a two-dimensional presentation may be performed for example using the principle of a self-organising map, known per se. In such map, one map unit may in this case represent the parametre set of an individual object, or a group of similar parametre sets. Adjacent map units on the map are close to each other also in the original multi-dimensional space in which the parametre sets reside, i.e. they relate to mutually similar objects as a whole. Grouping can also be based for example on the so-called K-means method or on some other known grouping method. Naturally, grouping and assigning the groups into quality classes can also be performed automatically by a computer, so that the visualization into a map is not required.
In a preferred embodiment of the invention, the groups are divided into quality classes according to the desired share of each quality class in the production. In this manner, it is easy to provide the specific proportions of the quality classes in the entire production.
In creating or changing the individual quality classes, the selection of a group to a quality class is preferably based on the sensory evaluation of a sample object which represents this group. By the sensory evaluation, it is possible to select such exemplary object to a specific class which bears the desired general impression. Assigning the objects which are the most similar in their general impression to the same class can then be easily performed by assign-
ing to this specific class groups that are close to each other in their mathematical distance.
A preferred embodiment of the invention comprises sorting of timber, for example boards. Evalua- tion of the quality of timber, for example sawn timber in particular, is typically based on evaluating a vast number of properties. Some of these properties, such as the objects' dimensions, derive from operations performed on timber. On the other hand, these types of natural goods have a vast variety of different properties which derive from the growth conditions and typically exhibit a large deviation. As a result, the final number of the quality classes is typically extensive. In such case of multi-dimensional parametre sets, the advantages of the invention in terms of the simplicity of creating the classification and adjusting it become evident. In addition to boards, the invention is naturally suitable for other sawn and/or planed sawn timber and also for logs that have not yet been processed.
Another preferred application of the invention is the use of the method for sorting stoneware, for example slabs of natural stone. In addition to properties deriving from the processing, stoneware also comprises properties which are characteristic of the raw material itself, and sorting according to these properties would be inconvenient and time- consuming with the traditional methods .
BRIEF DESCRIPTION OF THE DRAWINGS
In the following section, the invention will be described with reference to the accompanying drawings, in which
Fig. 1 represents an arrangement used in one embodiment of the method in accordance with the invention,
Fig. 2 illustrates one embodiment of the method in accordance with the invention, and
Fig. 3 represents one embodiment of the method in accordance with the invention as a flow chart.
DETAILED DESCRIPTION OF THE INVENTION
Fig. 1 presents an object 1 to be sorted which may be for example timber or stoneware. The fig- ure also presents part of a measuring system comprising a computer 2 and measuring devices 3a, 3b connected to the computer. The measuring devices are used for defining, for example optically, by means of ultrasound or by some other method which preferably does not involve contact with the object, different properties p of the object on the production line, and for combining these into a parametre set s (p a ,Pbr—rPi) which represents the properties of said object and is then used for the quality classification of the ob- ject. A number of objects are measured, after which the parametre sets are grouped into groups g (si,S2r...,S]c) of similar parametre sets, using the mathematical distances between the parametre sets in the i-dimensional space as the measure of similarity. Fig. 2 presents a two-dimensional map 4 which comprises map units 5. Each map unit represents one group g (sχ r s 2 ,.~, S k ) of parametre sets of similar objects. The two-dimensional map is created so that the original mutual mathematical distances between the pa- rametre sets and the respective groups are retained. In other words, proximity on the map 4 equals proximity also in the original i-dimensional parametre space. Most of the map units 5, i.e. groups g f have been divided into four different quality classes C. Over the centre area of the map there are some groups that have not yet been classified. These can be moved
into the desired quality classes by applying the proximity, i.e. similarity principle, so that at the end, each group will belong to some quality class. Boundaries of the quality classes can also be shifted later by changing the classes of the groups that reside on the border areas of the classes. The operation of creating the classification and changing it is illustrated in Fig. 2 by arrows between the groups and between the classes. Presence of a specific group in a specific class can be based for example on selecting the quality class by evaluating on the sensory basis an object which represents that specific group. Other similar groups can then be classified based on the proximity data of the map. Fig. 2 shows how in the method of the invention the boundaries between quality classes C are defined accurately, so that each parametre set s, with its group g which comprises similar parametre sets, clearly belongs to some of the quality classes. If the parametre sets were divided into quality classes for example by defining, in the parametre set space, central points of the classes, and by always determining the class border so that it would be in the middle of the central points of two classes, it would result in more inaccurate class boundaries, and classification of those parametre sets which reside close to the class borders would be more imprecise.
Map 4 of Fig. 2 may be based on the entire set of objects that are to be sorted, comprising, for example, a specific production lot. In such case, all objects to be sorted can be first measured, then the measured parametre sets can be grouped and divided into quality classes, and finally the objects can be physically separated from each other according to their groups and the classes of these groups. However, a more preferred alternative in many embodiments is to
first sort the objects into physically separate groups of similar objects on the base of the measurements, and to then assign each of the groups to a quality class. When there is a need for goods of a specific quality class, one can, in this manner, select objects from the groups which represent that specific class. Where necessary, the quality classification of the groups can naturally be later changed, over and over if so required. This mode of implementation necessi- tates only one step of physical separation of the objects, which makes the sorting process extremely efficient and flexible.
The sorting can also be based, in accordance with Fig. 3, on a specific, smaller sample lot which represents the entire set of objects that are to be sorted. The flow chart of Fig. 3 is an example of an embodiment of the invention in the timber industry. In the method, the desired parameter sets are first measured from the boards of the sample lot and grouped into groups based on their similarity. At least one representative for each quality class is then selected from the boards of the sample lot, for example by means of sensory evaluation, and the group of the pa- rametre set of that representative is assigned to the respective quality class. Next, the quality classes of the groups that are yet to be classified are determined based on the proximity of the groups. This can be performed manually, but also by a computer programme, by setting some conditions of distance for be- longing to the same class. Determining the classes manually is easy, in particular if the multidimensional parametre set is converted into a one-, two- or multi-dimensional map which retains the original mathematical distances of the parametre sets. When the classification of the sample lot is complete, it is used for sorting the boards that were
initially intended for sorting. In other words, for each board to be sorted, a parametre set which represents its properties is measured, and the location of this set is compared to the groups of the parametre sets of the sample lot. The quality class of the board is selected according to the quality class of that group in the sample lot which represents the board in the closest manner. Classification that has once been created based on the sample lot is later adjusted, where necessary, by moving groups from one class into another or by establishing new classes. The need for change may be caused for example by a reclamation issued by a customer concerning boards in a high-rated quality class. The method of the invention is not limited to the examples referred to above; instead, many variations are possible within the scope of the claims. In particular, the piece goods may be any natural, more or less reprocessed goods comprising various factors which contribute to the quality classification.
Next Patent: METHOD FOR DETECTION OF PRESENCE OF TARGET POLYNUCLEOTIDE IN SAMPLES