ELEVATOR DOOR MONITORING ARRANGEMENT FIELD OF. THE INVENTION The present invention relates to fault management of a computer-controlled door either in an elevator system or in another system containing the components in question. BACKGROUND OF THE INVENTION A mechanical system in normal operational condition comprises a certain amount of frictional force due to friction that resists movement. If the magnitudes of the frictional forces in the system can be determined by measuring or mathematically, this information can be utilized as an indicator of the operational condition of the system. An elevator system contains numerous components that are exposed to chafing and wear. The motion of the elevator car causes wear of components, including e.g. the elevator ropes and the guide rails of the elevator car. One of such components is the elevator door, which moves automatically on a horizontal rail. It is acted on by forces applied to it from different directions, and both its upper and lower edges are in contact with rails keeping the door movement on its track. There is also a frictional force opposing the motion of the automatic door. The operation of the door may be disturbed when a sufficient amount of dirt is accumulated on the door rail on the threshold of the elevator car. Due to this physical obstruction, the force opposing the motion of the door may grow to a magnitude such that finally the door control system is no longer able to open or close the door. The magnitude of the frictional force can not be measured directly. It is not possible to mount a separate "friction meter" on the door. The magnitude of the friction resisting the movement of the door has to be measured indirectly. It is possible to create a model of the system to be examined, in this case the elevator door, to study the forces applied to the door. One of the forces appearing in the model is the frictional force opposing the motion. Using the model, it is possible to calculate desired parameters when the magnitudes of the forces opening and closing the door are known and the acceleration or velocity of the door is measured. In this way, unknown parameters, such as frictional force, can be solved. Thus, the matter at hand is a problem of optimization and estimation of parameters. For example, in an elevator system the door assembly consists of a car door moving with the car and the landing doors on different floors. A modern automatic elevator door is opened and closed by means of a direct-current motor. The torque produced by the direct-current motor is directly proportional to the motor current. The energy of the motor is transmitted to the door e.g. via a toothed belt and the door moves on rollers. For reasons of safety, the landing door alone is closed without a motor by means of a closing device. The closing force of the closing device may be produced by a closing weight or a helical spring. The motor current and the corresponding torque are measured either from the door control card or directly from a motor current conductor. It is also possible to monitor a so-called tacho pulse signal of the motor. The tacho signal is a square wave whose frequency depends on the motor speed and therefore the door speed. The problem with prior-art solutions is that the frictional force acting on the elevator door can not be measured directly. This necessitates the use of an indirect method of estimating the magnitude of the frictional force. The magnitude of the frictional force is needed for an estimation of the time to failure of the door or for predicting a future time by which the operational condition of the door will decline to a level consistent with a given criterion. OBJECT OF THE INVENTION The object of the present invention is to detect the operational condition of an electric automatic door used in an elevator system or in some other system, by continuously monitoring the magnitude of the frictional force opposing the motion of the door. BRIEF DESCRIPTION OF THE INVENTION The method and system of the invention are disclosed in detail, hereinafter: Inventive embodiments are also presented in the description part of the present application. The inventive content disclosed in the application can also be defined in other ways than is done in the claims below. The inventive content may also consist of several separate inventions, especially if the invention is considered in the light of explicit or implicit sub-tasks or in respect of advantages or sets of advantages achieved. In this case, some of the attributes contained in the claims below may be superfluous from the point of view of separate inventive concepts. Within the framework of the basic concept of the invention, features of different embodiments of the invention can be applied in conjunction with other embodiments. The method of the invention can be used for real-time examination of the condition of an automatic door of an elevator or more generally an automatic door in a building. In more precise terms, an automatic door is a horizontally sliding door which is controlled by a motor and whose closing movement may be assisted by a closing device. The door is acted on by various forces, of which we are now particularly interested in the magnitude of the frictional force applied to the door. From the frictional force, it is possible to deduce an acute maintenance need and in less serious cases information regarding the frictional force can be used at best to anticipate a future time at which disturbances will most probably begin to appear in the operation of the door. The operational condition of the closing device of the door can be determined immediately. In an embodiment of the method of the present invention, the velocity of the automatic door is measured. This can be accomplished by using the so-called tacho signal obtained from the door motor. The tacho signal is a square wave in which the space between pulses depends on the speed of the motor and therefore on the door speed. The door speed can be calculated from the tacho signal. An essential part of the method is a dynamic model of the door. Some of the parameters in the model are updated after each pure door sequence. Pure door sequence means door opening and closing operations wherein no re-openings occur during the closing movement. The model includes the door and the closing device and the forces applied to these parts, including the frictional force. Using the model as an aid, the acceleration of the door is estimated, and from this the door speed as a function of time. The measured and the estimated instantaneous speeds are compared to each other and an error term is obtained. At each instant of time, the error term is a function of three variables (mass of the door, frictional force applied to the door, and force resulting from inclination of the door) . Next, the sum of the squares of the error terms is calculated, wherein each square of an error term is weighted by a desired weighting coefficient. For the so-called squared error term obtained as a result, a minimum value is found, in which situation the three model parameters being searched for are best in keeping with reality. From the magnitude of the frictional force thus obtained, the present state of the operational condition of the door can be deduced. In another embodiment of the method of the present invention, the acceleration of the door is measured using an acceleration sensor placed on the door. The method works as above except that in this case the quantity estimated in the dynamic model is acceleration. In the calculation of the error term, the instantaneous acceleration estimated from the model is subtracted from the instantaneous measured acceleration. In this embodiment, too, the error term is a function of the aforesaid three variables and the further processing for determining these parameters proceeds as in the example described above. The input parameters needed for the dynamic model of the door are door velocity, current of the motor driving the door, torque coefficient of the motor, motor friction and mass of door closing weight or force factor of closing spring. The calculation can be simplified by defining the mass of the door as a constant among the variables. In this case, the mass of the door is determined in connection with the start-up or commissioning of the system by taking the mean value from a desired number of door operations. The length of the "teaching period" to be examined may be e.g. about twenty door operations. Once the mass has been determined as a mean value of the results of the teaching period, the mass of the door is then set as a constant. After this, a function of only two variables (the frictional force of the door and the force caused by tilting of the door) is processed in the optimization logic, so the processing requires less calculation capacity and time than above. The mass of the door can be defined as a constant because it can be assumed that it will not change significantly in normal operating conditions. For immediate detection of a failure of the door closing device, it is possible to use a genetic algorithm (GA). Via the GA, both a correct door system model (with or without closing device) and unknown forces relating to door friction and tilt can be determined simultaneously. The parameters of the dynamic model of the door are coded into a chromosome of the genetic algorithm. In this connection, unknown parameters relating to the operation of the closing device, to the frictional force applied to the door and to the force caused by the angle of tilt of the door are genes, in other words, they together constitute a chromosome. The chromosome quality function is a squared error function, which can be regarded as an indicator of the performance of the solution or phenotype represented by the chromosome. With different gene values or alleles, correspondingly different phenotypes are obtained, of which the GA optimizer finally chooses, as a result of a search, a phenotype giving the minimum value. The gene values corresponding to this phenctype indicate the condition of the door system at the instant of examination. One of the advantages of the method according to the present invention is that the information relating the operation of the door can be saved. In this way, a data base covering the operating history of the door is created, on the basis of which it is possible to plan e.g. a suitable date ior the next maintenance. From the operating history, the present state of operation of the door can be deduced directly, and even the probability of failure and the need for maintenance at a future point of time can be predicted. LIST OF FIGURES Fig. 1 presents a dynamic model of an automatic door according to the present invention, Fig. 2 represents a method according to the present invention for determining the unknown parameters of the model, Fig. 3 represents another method according to the present invention for determining the unknown parameters of the model, and Fig. 4 represents a third method according to the present invention for determining the unknown parameters of the model. BRIEF DESCRIPTION OF THE INVENTION To determine the frictional force acting on the door, a dynamic model of the automatic door is created, wherein the forces applied to the door are observed. The dynamic model of the door is presented in Fig. 1. The basic law used here is Newton's second law, whereby the force applied to an object is obtained as the product of the mass and acceleration of the object. Another basic law relating to friction gives the magnitude of the frictional force opposing the motion of an object as the product of the coefficient of friction and the force (for an object sliding on an even surface, the force of gravity) pressing the object against the surface being examined. For the sake of clarity, in the dynamic model all moving masses are assumed to be concentrated on an individual mass point mdoor 10. Correspondingly, all frictional forces present in the system, except for the friction of the motor, can be combined into a single concentrated frictional force term Fµ,door- A model of the dynamic operation of the door system can be created using five different forces acting on it: force of the motor, force caused by the closing weight or spring, force caused by the angle of tilt of the door, internal frictional force of the motor, and frictional force caused by the door itself. The total mass of the system consists of the concentrated mass of the door 10 and the mass of a possible closing weight 11. All the moving masses comprised in the door mechanics are concentrated in the door mass 11. Fig. 1 shows the mass points and forces in the system as well as the positive directions of velocity and acceleration. From the dynamic model and Newton's second law, an expression for instantaneous acceleration adoor(t) of the door 10 is obtained: where Fmotor = B1 ■ 1motor(t) and Fcd(xd(t)) = mcd.g when the closing device is a weight and Fcd(xd(t)) = kcd . (xd0 + xd(t)) when the closing device is a spring. B1 is the motor torque coefficient, Tmotor is the motor current, Fmotor is the force caused by the motor, Ftilt is the horizontal component of the force caused by the tilt of the door, FCd is the force caused by the closing device, FµMotor is the internal frictional force of the motor, FµDoor is the concentrated frictional force acting on the door and caused by all the sub-components, mdoor is the common concentrated mass consisting of all the door masses and mcd is the mass of the counterweight. If the closing device is a spring, then mcd = 0. As a closing weight is more commonly used as a closing device, hereinafter we shall only deal with a closing weight. However, this does not restrict the device of the invention exclusively to a closing weight, but the closing device may be a mechanism that gets its closing force from a spring or some other arrangement. When samples of the quantities to be measured on the door are taken by means of the apparatus of the invention to determine the friction, this means a transition from the continuous-time world to discrete representation. In this case, (1) is changed to the form where instant t has been replaced by a sample taken at this instant with the running number k. Of the parameters of the dynamic model of the door, the mass of the closing weight, the torque coefficient of the motor and the internal frictional couple of the motor have to be known beforehand. The mass of the closing weight can be easily determined by weighing. The motor torque coefficient and the internal frictional couple of the motor can be determined by means of a dynamometer. Using a dynamometer, the motor torque can be measured as a function of motor current. The results obtained with different current values form an approximately straight line T, the equation for which is: where T is motor torque. By linear regression, the unknown quantities B1 and TµMotor can be determined as the slope of the regression line and its point of intersection with the y-axis. From the motor torque, the force acting on the door can be obtained by taking into account the power transmission mechanisms of the door system. In an example, the motor shaft carries a belt pulley of radius r, and a toothed belt running around the pulley moves the door leaves. In this case, the force moving the door leaves is easily obtained as Fmot0r = T/r. On the other hand, from the model it is possible to determine the unknown parameters, which in this connection are door mass, frictional force caused by tilt and frictional force acting on the door. Of these, the last mentioned parameter is the object of interest in a preferred embodiment of the present invention. A method according to the present invention for determining unknown parameters is presented in Fig. 2. The motion of the elevator door 20 is controlled by a control logic 26, from which a command to open or close the door is received. The door is driven by a direct-current motor, which is connected to a door control card. From this card, the motor current and a so-called tacho signal can be measured directly. The tacho signal is obtained from the motor's tacho generator, which detects the mechanical speed of rotation of the motor. In this embodiment, the tacho signal is typically a signal having the shape of a square wave. The frequency and pulse spacing of the square wave are directly proportional to the speed of the door motor and the velocity of the door. Between two successive pulses, the door always moves through the same sub-distance dx. The signals received from the control card and the commands given by the control logic are passed to a functional block 21 which takes care of collection and pre-processing, of information. In this block, the door motion data is filtered to remove from it those door opening operations during which the door has to be re-opened in the midst of the closing movement because of an obstacle, typically a passenger in the path of the door. During the period dt between to tacho pulses, the door moves through a constant sub-distance dx. In block 21, it is now possible to calculate the velocity vd of the door at each instant k of time: The pre-processing block also calculates weighting coefficients for later calculation of an error term. By- using weighting coefficients, desired error terms can be weighted more than others. In the pre-processing block 21, all the information relating to door opening and closing operations is combined for further processing. The next step in the method is processing of the dynamic model 22 of the door. The model was described above and illustrated in Fig. 1. As stated above, the input parameters fed into the model are motor torque coefficient, frictional couple of the motor, mass of door closing weight, motor current, period of time dt and velocity vd of the door. In the model, the acceleration of the door is estimated as a function of four variables as follows. where ΣFk(.) is the sum of the forces acting on the door at instant k. From the estimated acceleration of the door, the velocity of the door can be estimated as follows: where vd,0 is the velocity of the door at instant t=0. In the next step, the estimated velocity of the door and the door velocity calculated in the pre-processing block are passed into a differential block 23. The estimated instantaneous velocity is subtracted from the measured instantaneous velocity, producing an error term ek as a result. The error term ek is a function of the three variables md, Fµ and Ftilt. By applying weighting coefficients Wi, a so-called squared error term E can now be calculated in block 24: In the next step in the block diagram of the method of the present invention, the squared error term E is transferred to an optimizer 25. The function of the optimizer is to minimize the function (7) of the three variables. When the minimum value is found, the variable parameters corresponding to this have been estimated for the mass of the door, the frictional force opposing the motion of the door and the force caused by the tilt of the door. In the examples illustrated in Fig. 2-4 and in the model in Fig. 1, it is possible to define one or more of the force parameters in the model as constants if it is desired to simplify the model and the calculation under certain assumptions. Fig. 3 presents another example of the method of the invention for detecting a failure of an automatic door. The operation in this example is very close to the method presented in Fig. 2. The control logic 36 of the elevator system issues an opening or closing command to the door. In the case of elevators in which no motor tacho signal is available, the motion of the elevator door must be observed by other methods. One method is to mount on a door leaf 30 an acceleration sensor to monitor the acceleration of the door. The measured acceleration ad is passed to a block 31 for collection and pre-processing of information. As in the above-described block 21, the door motion data is filtered to remove from it those door opening operations during which the door has to be re-opened in the midst of the closing movement because of an obstacle in the path of the door. After this, the velocity vd of the door is calculated in block 31 from the following basic formula: where Vd,0 is the initial velocity of the door at instant t=0. In other respects, the pre-processing block 31 works like the pre-processing block 21 in Fig. 2. The signals between block 31 and the dynamic model 32 of the door are consistent with the method of Fig. 2 with the difference that the error term E is calculated from acceleration values instead of velocities. In the model 32, an estimated door acceleration is calculated from equation (5). This information is fed directly into a differential block 33, where the measured acceleration, which in this case is obtained from the sensor, and the estimated acceleration obtained from the model are subtracted from each other. This produces an error term e, which is a three-variable function of the same type as in the example in Fig. 2. The error is squared with desired weightings in block 34 in the way described above. Correspondingly, optimizer 35 works in the same way as optimizer 25. As a result, the same three unknown parameters are obtained as above. In an embodiment of the model, the three unknown parameters of the model are determined once in conjunction with the start-up of the system. To ensure the accuracy of the parameters, several door operations are needed for each floor. A suitable estimate for the number of door operations is at least ten. When the system is subsequently in its operating condition, the previously defined model of the system is in use and this makes it possible to compare the existing model to recently collected new information about the motion of the door. After the comparison it is possible to conclude e.g. whether the frictional force Fµ has changed significantly. A clearly increased friction between the door and the door rail is quickly detected from the error terms ek, i.e. from the residuals of the model. The residuals of the model can be e.g. analyzed statistically. It is possible to evaluate e.g. the mean value, variance, distortions of distribution, and number of peaks. The error term can also be analyzed in respect of frequency range. By these methods of analysis, it is possible to determine characteristics typical of different failure situations. For example, an increase of the friction opposing the motion of the door will appear as a deviation of the mean value of the residuals from zero. For an analysis of failure type from the statistical quantities or the frequency range signal it is naturally required that failure types can be clearly distinguished from each other and from an error-free operating condition by examining the amplitudes and frequencies of the spectrum components. This may be difficult. In another embodiment of the model, an analysis of the operational condition of the door can preferably be performed each time the door is closed or opened. The method in this case is one of continuous detection. The processing and analysis of the collected information have to be carried out within the period of time between two door operations. In the case of an elevator, this processing period should be of the order of max. 15 seconds, which is the time needed by the elevator in a driving cycle between two successive floors. Of course it is not absolutely necessary to include every door operation in the analysis. Therefore, it does not matter if the analysis of one door operation should take more time than about 15 seconds as stated above. In this case the efficiency of fault diagnosis is naturally impaired. Even if not every door operation is included in the analysis, it is still important to count the number of all floor- specific door operations. This is an essential item of information when in the event of a failure the average useful life of the door is to be determined. The analysis performed by the optimizer can be simplified by assuming the door mass to be constant. Anyway, the door mass has to be defined in connection with the start-up of the system. In practice, the model is given a constant door mass value which is determined e.g. as an average of the mass values obtained from the first 20 door operations at each floor. After this "teaching period", the function of the optimizer is to find values for two unknown parameters, the friction opposing the motion of the door and the force caused by the tilt of the door. The amount of computing work is now reduced and the search for parameters becomes easier. After the teaching period, the method in this example of the present invention works like the method presented in Fig. 3, with the difference that ma is now a fixed constant parameter and that both ek and E are functions of two parameters. A typical door failure situation is for example a fault occurring in the bearing of a roller guiding the door, preventing smooth sliding of the door on the roller. In such a situation, the frictional force Fµ of the door mechanism increases either abruptly or slowly with time, depending on the nature of the failure. One possibility is to determine from this information the need and time for maintenance. Anothex possible type of fault is a failure of the door closing device. Such a fault may arise e.g. when the closing weight has been removed in connection with maintenance and the serviceman has forgotten to mount it again. A failure may be due to the wire cable of the closing weight being broken. Such a fault appears as a sudden and large increase in the force Ftilt caused by the tilt of the door. It can be inferred that such a large tilt of the door is not due to a real tilt but to a disappearance of the closing force. In this connection there arises a need to automate the process of inferring the operational condition of the closing device by a suitable method. Genetic algorithms can be utilized for this purpose. Using these algorithms, it is possible to determine both the right door model (in which a closing device is either included or not) and the unknown forces FµDoor and Ftilt. While searching for the forces of friction and tilt, the genetic optimizer simultaneously finds the system model that produces the smallest force of tilt. Genetic algorithms are based on the principle of creating an artificial evolution by using the computing logic of a processor. The question at issue is how to obtain as advantageous a final result ("phenotype") as possible by varying the properties of a "population". In the process of variation, the genetic operations used are "selection", "hybridization" and "mutation". The strongest members of the population "make it", and the properties of these ones are passed on to the next generations. In an example of the method of the present invention, the population is a number of model parameter vectors. In this connection, one parameter vector corresponds to one chromosome. Each chromosome has genes. Each gene in this connection corresponds to one of the model parameters to be estimated, which now are operation of the closing device, frictional force of the door, and force of tilt of the door. These three genes together can be called a phenotype. The operation of the genetic algorithm is such that first a population is created with gene values selected at random. For each chromosome in the population, an "efficiency" or a quality value is calculated, which in this example is the above-described squared error term computed from the dynamic model of the door. In the genetic algorithm, the search proceeds generation by generation. From each generation, the most efficient chromosomes, i.e. the ones that give the lowest squared error term value, are selected and included in the next generation. From the best alternatives after this selection, the next generation is created via hybridization and mutation. As a result of the genetic operations, a new kind of population is obtained in which the genotype of the chromosomes differs from the earlier population either completely or only in some of the genes. For the new generation, an efficiency, i.e. squared error terms are calculated, and a chromosome having the best efficiency is again obtained as a result. After this, the sequence of numbers of the squared error terms is checked to see if it converges and if a sufficient number of generations have been processed to guarantee convergence. As a final result, the genes of the best individual in the last generation reveal the magnitudes of the unknown forces and the operational condition of the closing device. The operation of the above-described genetic algorithm can be combined with both of diagrams 2 and 3. Diagram 4 represents the operating principle by way of example when the genetic algorithm is combined with diagram 2. In the automatic door 40, the current of the door motor and the tacho pulse signal of the motor are measured. In the pre- processing block 41, the door velocity is calculated, and the result is passed to the differential block 43 and to the model 42 of the door. In this example the door mass is assumed to be constant. In the model, the door velocity is estimated and likewise passed to the differential block 43. A squared error term calculator 44 and a so-called GA optimizer 45 form a loop, whose operation was described above in connection with the description of the genetic algorithm. The information about the genes is transferred from the GA optimizer 45 to the error calculator 44 and correspondingly the efficiency value, i.e. the squared error term E is passed from the error calculator 44 to the GA optimizer 45. As a final result of the search, the optimizer gives parameters CD, FµDoor and Ftilt. CD means the operational condition of the closing device, where e.g. the value one may represent error-free operation of the closing device and zero a failure of the closing device. These three parameters are returned back to the model, so the model takes the performance of the closing device immediately into account. Thus, in addition to the force parameters, the model that best describes the system is found immediately. The door opening and closing commands come from the door control system 46. The dynamic model of the door is now where the term CD is one when the closing device is in operation, and CD is zero when the closing device does not work. In order that the genetic algorithm should be able to find the system model that produces the smallest tilt angle, the force of tilt Ftilt is also included in the error function where K is a scaling coefficient, G is the sequential number of the generation in the genetic algorithm and G1 is a limit value for generation G after which the force of tilt is no longer included in the error function (10). The result of this arrangement is that in the early stage of the search, when G