WO2011048231A1

WO2011048231A1 - Intelligent device and method for compensating for ram sag in machine tools

Info

Publication number: WO2011048231A1
Application number: PCT/ES2009/070453
Authority: WO
Inventors: Andrés BUSTILLO IGLESIAS; Maritza Correa Valencia; Rodolfo Elias Haber Guerra
Original assignee: Universidad De Burgos; Consejo Superior De Investigaciones Científicas
Priority date: 2009-10-23
Filing date: 2009-10-23
Publication date: 2011-04-28
Also published as: ES2398814B1; DE112009005232T5; ES2398814A1; CN102483622A

Abstract

The present invention consists of an intelligent device and the method using said device in order to compensate for ram sag by using probabilistic calculation, specifically by means of Bayesian networks, and internal model control, specifically an adaptive neuro-fuzzy control device.

Description

SMART DEVICE AND PROCEDURE FOR THE COMPENSATION OF BUTCH FALLS IN MACHINES

TOOL

OBJECT OF THE INVENTION

The present invention is encompassed in the field of machine tools and specifically in those consisting of a ram, such as milling machines.

Said invention is an intelligent device and the method using said device to compensate for ram falls through the use of probabilistic calculation, in particular by means of Bayesian networks, and an internal model control, specifically a neuro-fuzzy adaptive control device. .

BACKGROUND OF THE INVENTION

Rams in machine tools, in spe ^¬ cial milling, horizontal ram, which is known as machines in column configuration, its ^¬ fren by definition a geometric axis error (typically the Y - axis) defined by This element is ^¬ tructural. This error is due to the flexion of the ram due to its weight and the weight of the machine tool head, especially milling, which they incorporate.

Throughout this document the axes of the machine tool are considered as follows: X axis, is the longitudinal axis of the bench; Y axis, is the longitudinal axis of the column; Z axis is the longitudi nal axis ^¬ ram. This problem is exacerbated today because of the proliferation of machine tools incorpo ^¬ ran over a head thanks to an automatic change of heads. Each head has a different weight and therefore the flexion that occurs is different.

The traditional method to prevent this bending is known as "electronic compensation" is ^¬ gram in the computer numerical control (CNC) milling machine compensation table axis Y of the machine to compensate this fall of ram . The problem with this solution is that it does not allow to recover the loss of orthogonality between the axes generated by this fall. Thus, for mechanical grinding heads for milling machines with which large plate milling cutters are used, this deviation in planning operations can produce steps between each consecutive planning of up to 0.5 mm.

To solve this problem, currently some manufacturers of machine tools when they perform the final machining of the rams introduce tensors inside them that simulate the weight of the head. When unscrewing these tensioners after their manufacture, the ram presents a curvature in its longitudinal axis that compensates for its subsequent fall. This solution is only valid if the machine tool incorporates a single head. If the machine incorporates several heads of different weight this compensation is usually made for the heaviest or for the one that is supposed to be more used, according to the client's own criteria. To offset the remaining heads electronic compensation is used to ^¬ quently described or fixed tensioners are introduced into the ram to which different tension is programmed according to the head that incorporates the machine at all times.

In this traditional problem impor ^¬ tant development in recent years of active compensation systems vibration machines TOOLS ^¬ tas, especially milling, which should be incorporated as close as possible to the tool is added, that is, in the ram of the machine tool. These systems should be placed on the upper / lower surface of the ram to compensate for vibrations in the X and / or Z axes of the machine and on the lateral surfaces of the ram to compensate for vibrations in the Y axis of the machine. Therefore, introduce new weights and modifi cations ^¬ local stiffness should be compen ^¬ sadas. In addition, depending on the use of the machine tool, the range of forces that the active vibration compensation system must compensate for will be different, so its weight, mooring surface size and location will also be different.

The most logical to compensate these modi fications ^¬ method is to perform a Finite Element Modeling (FEM) of the ram to estimate its rigidity and calculate the voltage tensioners which engage during machining of the ram. But, carrying out a study of this nature for each ram of machine tool to be manufactured is too costly economically ^¬ in the case of large machine tools that incorporate high customization in dimensions, heads and optimization to specific processes, especially in the case of milling machines

That is why the development of a device and the corresponding procedure that can perform a sufficient estimate of the tension to be applied to the tensioners that are coupled during the mechanization of the meat ^¬ ro automatically becomes an application of industrial interest.

DESCRIPTION OF THE INVENTION

The present invention is charac ^¬ terizada established and in the independent claims, mien ^¬ after the dependent claims describe other characteristics of the same.

In view of the above statement, the present invention relates to a Inteli device ^¬ people for compensation falls ram in machine tools comprising the following ele ^¬ ments: a machine tool with computer numerical control (CNC), means for the application of probabilistic calculation and means for control by internal model.

Also, the present invention relates to a method for compensating ram falls on machine tools using an intelligent device according to claim 1 comprising the following steps:

- device start by the CNC,

- Acquisition of values for the variables to be used, - Use of probabilistic calculation to estimate the tension of the compensating cylinders using the values of the variables by the means for the application of probabilistic calculation,

-election of the tension of the compensating cylinders, -modification of the machining program by the means for internal model control that implements an internal model control (BMI),

-implementation of the real machining conditions by the CNC.

BRIEF DESCRIPTION OF THE FIGURES

Herein, with a set of drawings, illustrative example prefe ^¬ ent, and never limiting the invention it is complemented.

Figure 1 represents a perspective of a machine tool.

Figure 2 represents the structure of a Bayesian network augmented by tree of the variables.

Figure 3 shows a diagram of smart vo ^¬ dispositi showing paths exchange information between elements forming said dispo ^¬ sitivo.

DETAILED EXHIBITION OF THE INVENTION

In the present embodiment of the invention, an intelligent device is described and the procedure carried out therein by means of a calculation system that allows to achieve the optimum compensation of the ram's fall (1.1). Said calculation system is based on a distributed system and prediction calculation ^¬ ten sion of compensating cylinders and control machinable Zado.

The elements forming the device intelli ^¬ people are:

-Machine tool with its Computer Numerical Control (CNC) (1), specifically milling machine, which mechanizes the ram (1.1).

-Media for the application of probabilistic calculation (2), specifically a workstation with Bayesian Networks application.

-Media for control by internal model (3), specifically a neuro-fuzzy adaptive control device.

-Network medium for communication between the elements that make up the smart device.

The machine tool, specifically milling machine, is responsible for the final milling or finishing of the outer surface of the ram (1.1) that will be incorporated into other milling machines. On this outer surface both the ram guidance system (1.1) and its drive will be supported, so that this surface determines the final geometry of the new milling machine, specifically the movement of the Z axis in column machines.

The CNC (1.3) is responsible for starting up the smart device by means of a virtual key programmed in the programmable logic controller (PLC, acronym of the English name "Programmable Logic Controller") (1.2) of the machine. This virtual key must be pressed by the operator when a ram (1.1) has been placed for final machining on this machine. Then the virtual key sends to the workstation (2) the machining program from which extract the values of the variables preci ^¬ sa for calculating voltage cylinders com ^¬ thinkers and so starts the intelligent device presented here.

In addition, the CNC (1.3) is responsible for, during the machining of the ram (1.1), modify in real time the machining conditions according to the variation proposed by the neuro-fuzzy adaptive control device (3).

This modification in real time condi ^¬ tions machining by the CNC (1.3) is possible since it is responsible for controlling the herra ^¬ lie follows the programmed path in the machining program using sensor signals inter ^¬ position and others available to the machine; In addition to the CNC (1.3), the milling machine has a programmable controller or programmable logic controller (PLC) (1.2) whose programming is more accessible than that of the CNC (1.3), usually closed or restricted by its manufacturer.

After starting the intelligent device CNC (1.3) receives from the workstation (2) ten ^¬ sion (TCC) the operator must apply to cilin ^¬ compensators ders before machining and sample the operator in his screen. If the workstation (2) cannot discern the appropriate voltage, the CNC (1.3) shows the operator the probabilities for each type of CBT so that it is he who makes the decision of the voltage to be applied.

The workstation (2) with Bayesian Networks application provides the voltage value to apply to compensating cylinders (TCC) based on known variables. This value allows indi ^¬ TCC uprightly obtain the appropriate curvature of the ram (1.1) to be machined, as removing the compensating cylinders after machining has previously applied the appropriate voltage, the ram (1.1) curves only.

The prediction of CBT is carried out based on the value of 9 variables that we call predictor variables. The value of the predictor variables is obtained by the application of both the ram machining program (1.1) included in the CNC (1.3) of the machine and the Database (4) of Company Orders.

These predictive variables are:

• Maximum length of the ram (1.1) in its posi ^¬ outermost (LMC).

• Critical weight of the head or head to compensate

(PCC).

• Dimensions of the active shock absorber (DAA).

• Forces to be applied by the shock absorber (s) / s (FAA).

• Weight of the active shock absorber (PAA).

• Existence or not of fixed internal cylinders of compensation of the fall for secondary heads (CIF).

• Tension value of fixed internal compensation cylinders (TCIF).

• thickness of the plates that make up the des ^¬ pare the ram (1.1) (ECC).

• Mooring position of the active shock absorber

(PAAA).

The equation, which defines the value of CBT has the following form: CBT = f (predictor variables)

The network structure is built on the 9 predictor variables obtained before starting the machining process and a tenth variable, the CBT, called the class variable. The 9 variables used in the network are characteristics of the manufacturing process, the design of the ram (1.1) and the subsequent operating conditions of the ram (1.1) to be machined.

To generate data for the training of the network, the Finite Element Method (FEM) has been used.

Since the TCC is continuous variable defi nition ^¬, while variables in a Bayesian network must have a discrete number of states, this variable is discretized according to the attached table below values.

A Laplace correction is applied to the Bayesian network to thereby assign a proportion of participation to cases that are not present at the time of training but may appear in the future. To describe the procedure for generating the Bayesian network, we will use the following terminology and notation:

Event or state of nature i (xi) ^¬ tuación is whether in the future on which we want to know certain information.

A priori probability of a state x (p (xi)): probability that a state of nature i occurs from the initial information, that is to say without any evidence.

Study or sample information: this is usually the extra information that can be obtained after conducting a study or test. The results of a study or sample can be represented by different res ^¬ indicated. The set of findings is called evidence and is usually noted as e.

Conditional probability of xi given e (p (xi | e)): It is the probability when an event A influences the result of a second event B.

Probability Aposteriori of xi given e (p (xi | e)): It is the probability of an event after it has occurred. If P (xi | e) represents the aposteriori probability, then P (xi | e) = P (e, x ±) | P (e).

In the Naive Bayes method, the probability that a k-th example belongs to the ith class of the variable xi is calculated according to the following equation:

The algorithm called ^¬ do in English "Tree Augmented Network" (TAN) proposed by Friedman has been used for the generation of the Naive Bayes structure augmented to tree. In this algorithm the amount of mutual information conditioned to the variable is taken into account class. The amount of mutual information between various discrete bles ^¬ X and Y conditional variable C is defined by the following equation:

To apply have followed the following pa ^¬ sos:

Step 1. Calculate I (Xi, X _j | C) with i <j; i, j = l, ..., n.

Step 2. Construct a complete non-directed graph whose nodes correspond to the predictor variables: Xi, ..., X _n . Assign each edge by connecting the variables Xi and X a weight given by I (Xi, X _j | C).

Step 3. Assign the two edges of greatest weight to the tree to be built.

Step 4. Examine the next edge of greater weight, and add it to the tree unless it forms a cycle, in which case the next edge of greater weight is discarded and examined. Repeat step 4 until seleccio ^¬ nar n-1 edges.

Step 5. Transform the tree undirected resul tant ^¬ one run, choosing a variable as root, to then direct the other edges.

Step 6. Build a TAN model by adding a no ^¬ do labeled as C and then an arc from C to each predictor variable X ±.

The TAN structure obtained for the ^proposed problem is illustrated in Figure 2.

For the validation of the Bayesian Network, the cross-validation method or K-fold cross-validation was used, according to its English name, and a file with data Experimental In the cross-validation method, the initial data set is divided into K subsets, of the K subsets a single subset is saved as validation data to test the model, and the other Kl subsets are used as training data. The process is repeated K times, partitions, where each of the K subsets is used exactly once as validation data. The results of the parti tions ^¬ averaged, or otherwise combined to produce a single estimate of the accuracy of classi ^¬ er, K = 10 was taken.

The output of the model after validation is a contingency table or confusion matrix, which summarizes the results of data well and poorly classified ^¬ two for each state of the class and a general precision value that indicates the percentage of success with that the model will classify the new data presented to the network. After validation, the application is ready for use, questions can be asked to the Bayesian network on different types of questions to find the posterior probability of the different possible answers.

In our case, the type of question is the razo ^¬ tioning predictive or causal inference where the prediction of effects sought. The Bayesian network is asked "What is the probability of each state of the TCC class given certain manufacturing requirements?" The manufacturing requirements will be the variables LMC, PCC, DAA, FAA, PAA, CIF, TCIF, ECC and PAAA already defined above.

In a specific case, the Ba ^¬ yesiana network would be asked about the probability of CBT according to the following values: P (TCC | LMC = 1, PCC = 500, DAA = 150, FAA = 300, PAA = 40, CIF = 0, TCIF = 0, ECC = 15 and PAAA = 250). By propagating this evidence, the network calculates the following probabilities of CBT, expressed here as both by one: low voltage 0.02; average tension 0.89; medium-high voltage 0.09 and high voltage 0.00. With these requirements the states of the class with higher probabilities identify the best voltage to apply that corresponds to the average voltage since its probability is 89%. This reasoning is correct since it corresponds to values obtained in the experimental tests.

If one of the TCC states is predominant, probability greater than 80%, it is transferred to the CNC (1.3) of the machine, if it is not, the probabilities ^¬ des for each state of the TCC class are warned to the operator for To analyze the situation.

The neuro-blurred adaptive control device (3) is the device that receives information from the machine's internal sensors (1.3) or, additionally, from external sensors installed in it. In addition, it monitors and modifies variables of the CNC (1.3) related to the execution of the part program such as the speed of advance and the speed of rotation in order to optimize the machining time and minimize the error due to the flexion of the ram (1.1 ).

This adaptive control device is a computer or an embedded control system connected to the CNC (1.3), in the specific example proposed here by a network medium, which sends the desired part program to the CNC (1.3) run. In addition, this device receives information from the sensors included in the machine tool with its CNC (1) through the network medium. This device implements a control mode ^¬ internally (BMI, which stands for the English name "Infernal Control Model") based on neuro- fuzzy controllers for, from the tension applica ^¬ car is due to compensating cylinders and real-time machining conditions, modify the speed of advance or rotation of the milling machine that mechanizes the ram (1.1), that is, modify the machining conditions provided for in the machining program, sending the new conditions to the CNC for implementation .

The controller is based on fuzzy logic and corpora ^¬ precontrol module based on networks neuron Ies (neuro-fuzzy controller). According to preferred embodiments of the invention, the parameters defining the neuro-fuzzy controllers are DETERMI ^¬ nan so that minimizes an Index of merit.

Blurred logic is based on fuzzy sets, whose elements are associated with a membership function that indicates to what extent the element is part of that fuzzy set. Forms functions perte ^¬ nence most typical are trapezoidal, triangular and siana gaus-. That is, the fuzzy logic is based on heuristic rules of the form IF (antecedent) THEN ^(¬ sequent), where the antecedent and consequent are also fuzzy sets, either pure or result of operating with them. The rules that determine the rennovation ^¬ nence of the elements of fuzzy sets is based, in the case of the ram (1.1) in the designer's experience and on information provided by the Bayesian networks.

The network medium is responsible for joining the three allowing the elements described above comuni ^¬ cation therebetween and minimizing the delay in communications between them.

Said network medium can be a field bus, belonging to the Profibus family, or others depending on the type of network: mpi, ethernet, internet, etc. In any case, the network medium is characterized by the maximum delay that it introduces in the transmission of monitoring and control signals between the CNC (1.3) and the control medium, which will be called L _mec iio re _d - The maximum global delay of the communications process L _max will therefore be the sum of the delay introduced by the network medium L _{medi0 network} and the delay due to the dead time of the process L _carner0 , that is, L _max = L _carner0 + L. This delay, in preferred embodiments of the invention, is less than 60 seconds.

A preferred use of the network medium for the probabilistic calculation allows several elements to work at the same time for the execution of said calculation, so that the calculation time is minimized.

Claims

1. Intelligent compensation ram falls in machine tools com ^¬ device turns the following elements: a machine tool with CNC (1), means for applying probabilistic calculation (2) and means for internal model control ( 3) .

2. Intelligent device for compensating ram falls on machine tools according to claim 1, characterized in that the communication between the elements that form the device is through a network means.

3. Method for compensating ram falls in machine tools using a smart ^¬ dispo sitivo according to claim 1 comprising the following steps:

- Start-up of the device by the CNC (1.3), - Acquisition of values for the variables to be used, - Use of probabilistic calculation to estimate the tension of the compensating cylinders using the values of the variables by the means for the application of probabilistic calculation (2),

-election of the tension of the compensating cylinders, -modification of the machining conditions foreseen in the machining program by the means for control by internal model (3) that implements an adaptive neuro-blurred control,

-implementation of the new machining conditions by the CNC (1.3).

4. Procedure for compensating ram falls on machine tools according to claim - ^¬ tion 3 characterized in that the device is started by the CNC (1.3) using a virtual key of the programmable logic controller (1.2).

5. Method for compensating ram falls in machine tool according to claims 3 ^¬ characterized in that the values of the variables are acquired machining program and a database (4).

6. Method for compensating ram falls in machine tool according to claims 3 ^¬ wherein the variables used are selected among the following: maximum length of the ram at its outermost position, CML; critical head weight or head to compensate, PCC; active damping dimensions, DAA; forces to be applied by the shock absorber (s), FAA; weight of active shock absorber, PAA; existence of fixed inner cylinders of fall compensation for secondary heads, CIF; tension value of fixed internal compensation cylinders, TCIF; thickness of the sheets that make up the ram's walls, ECC; mooring position of the active shock absorber, PAAA.

7. Method for compensating ram falls in machine tool according to claims 3 characterized ^¬ for probabilistic calculation tico Naive Bayes one network is used.

8. Method for compensating ram falls in machine tool according to claims 7 ^¬ characterized Naive Bayes network has been trained using the finite element method.

9. Method for compensating ram falls in machine tool according to claims 7 characterized ^¬ Naive Bayes network is increased to tree algorithm using the Friedman.

10. Procedure for compensating ram falls on machine tools according to claim 9, characterized in that the raised tree network is corrected using the Laplace correction.

11. Procedure for compensating ram falls on machine tools according to claim 7, characterized in that the Naive Bayes network is validated using the cross-validation method and experimental data.

12. Procedure for compensating ram falls in machine tools according to claim 3, characterized in that the device is selected by the device if the probability is greater than 80% and the operator performs it if It is equal to or less than 80%.

13. Procedure for compensating ram falls in machine tools according to claim 7, characterized in that the value of the tension of the compensating cylinders is discrete.

14. Method for compensation falls ram machine tool according to claim 3 characterized in that the neuro-fuzzy adaptive control modifies condicio ^¬ tions machining in real time to adapt veil ^¬ cidades forward and spindle.

15. Procedure for compensating ram falls on machine tools according to claim 3, characterized in that the network means is used for the probabilistic calculation so that several elements work simultaneously for the execution of said calculation.