ES2293845B2 - SYSTEM OF MONITORING AND CONTROL OF PROCESSES OF SURFACE THERMAL TREATMENT OF MATERIALS WITH LASER THROUGH AN ADAPTIVE NEURONAL CONTROL BY MEANS OF REFERENCE MODEL. - Google Patents

Abstract

System for monitoring and controlling processes of surface heat treatment of laser materials by means of an adaptive neuronal control using a reference model. It consists of a real-time monitoring and control system of surface heat treatment processes of laser materials by means of a neuronal control using a reference model, which uses as input variable the maximum temperature on the surface of the material and as a performance variable the power output of the laser source. The system cyclically calculates the actuation signal on the power of the laser source (8), based on the tracking error (4). The control signal on the power of the laser source (8) is calculated in two stages: first, the proportional, Integral and Derivative linear controller, PID (5) generates, based on the tracking error (4), the corresponding action (6), and subsequently, from said signal and the state of the process (2), the neuronal controller by reference model (7) calculates the action signal on the laser source (8).

Description

Process monitoring and control system of surface heat treatment of materials with laser by an adaptive neuronal control by reference model.

Technical sector

The invention falls within the technical sector of surface heat treatment processes of materials by laser technology, focusing on monitoring and control advanced of these processes.

State of the art

In recent years, laser technology is being successfully applying in a wide variety of heat treatments superficial, providing enormous flexibility and speed, allowing localized surface treatments and very small thicknesses, with minimal material alteration base or substrate, which has given it a certain sense of technological maturity, although an in-depth analysis allows detect still countless aspects and unresolved issues, mainly in the area of monitoring and control automatic process.

Among the main limitations for the Large-scale industrial application of laser technology fits highlight the possible lack of uniformity in treatment as consequence of the high sensitivity of the process to external disturbances, essentially derived from instabilities in the laser source and imperfections in the surface finish of the item to be treated

Irregularities in treatment may give result in unacceptable behavior of the final product, such as consequence of a decrease of its resistance to wear, to fatigue or corrosion, requiring postprocessing thereof or even entail its rejection and destruction, with the consequent cost increase

These considerations allow establishing the need to design a monitoring and control system for real-time process capable of keeping it within nominal working conditions, improving the uniformity of treatment and increasing the final quality of the product obtained.

While in the literature you can find various references to control systems for processes treatment of materials with laser, which can be cited as examples EP0836905, EP0086540 and JP2003183726, most of the references are limited to linear control systems, essentially Proportional, Integral and Derivative type controllers Error (PID) and some controller through feedback state vector, with no references found about of the application of a neuronal control system using a model of reference to material treatment processes with To be.

In any case, despite the high non-linearity and complexity of laser material treatment processes, its theoretical foundations have been widely studied in recent years, and extensive documentation can be found in the literature, including various mathematical models of high performance, some of the most significant developed by the patent applicants (JL Ocaña et al . "Three-dimensional Numerical Model for the Simulation of Processes of Surface Treatment of Materials with Laser". Metallurgy Magazine. 35. 75-83. 1999 .). These models are mostly based on finite elements and are very useful for the study and predictive analysis of treatment.

However, as a result of its complexity and high calculation time, these models are not suitable for real-time treatment simulation, being necessary resort to state-of-the-art nonlinear modeling techniques, such as modeling using artificial neural networks to get a proper treatment model and fast enough for direct real-time evaluation.

As a result, they have not been found references to integration in monitoring systems and surface heat treatment process control of laser materials of a stage of process monitoring that verify the consistency between the information coming from the sensors with the one provided by a high neuronal model benefits incorporated into the monitoring and control system of the process, revealing the originality of the system Patent monitoring and control.

Brief description of the figures

Figure 1 presents a block diagram of the proposed monitoring and control system in its mode of complete realization, in which the monitoring system and control cyclically calculates the action signal on the output power of the laser source (8) based on the signal of tracking error (4), defined as the difference between the desired maximum temperature value, specified by the setpoint of reference (1) and the value actually achieved by the treatment (2), value that has been previously filtered and conditioned (12) and its validity verified by the process supervision stage (13).

       \ newpage
    

In order to optimize the benefits of process monitoring and control system, the system control has been divided into two stages: a linear controller type Proportional, Integral and Derivative, PID (5) which generates the corresponding actuation signal (6) based on the error of tracking (4) and secondly a neuronal controller adaptive by reference model (7), which calculates the actuation signal on the output power of the laser source (8) from the actuation signal of the PID controller (6) and the process status, represented by the maximum temperature value on the surface of the material really reached by the treatment (2).

Figure 2 describes in a detailed way the operation of the PID linear controller (5), which calculates the actuation signal (6) as a linear combination of three actions basic control: the proportional action (15), that is a action proportional to the following error (4), the action integral (18), an action proportional to the temporal integral of the follow-up error and the derivative action (24), a performance proportional to the temporary derivative of the tracking error (22). In order to narrow the range of action, the PID controller (5) incorporates an anti-saturation mechanism (26).

Figure 3 presents a detail of the Neural controller operation by means of reference (7), which calculates the action signal on the output power of the laser source (8) from the signal of PID controller performance (6) and process status, represented by the maximum surface temperature value of the material actually reached by the treatment (2).

Figure 4 presents a detailed outline of the functioning of the incremental neuronal model of the process incorporated into the supervision stage (13), which estimates the temperature increase at the next instant of sampling (30) depending on the maximum temperature on the surface of the material estimated by the model at the current time (28) and the performance on the output power of the laser source (8) in the instant current. The maximum temperature value at the current instant (28) is obtained by adding the ambient temperature value (35) to the integral of the temperature increase (34), calculated by accumulation.

Figure 5 presents a simplified scheme of operation of the supervision stage (13), which verifies the coherence between the maximum surface temperature signal of the material already filtered and conditioned (12) with the value of estimated temperature using the incremental neuronal model of process (28), comparing the difference between both values, expressed in relative value (40) with an acceptance threshold or rejection (41), defined based on the particular characteristics of the heat treatment processes of laser materials and the type of temperature sensor used.

In the event that the difference between the two values, expressed in relative value (40) exceed the threshold value of acceptance (41), defined as the percentage of difference admissible, the process supervision stage (13) emits a signal alarm, allowing the system operator to take measures timely corrections.

Figure 6 presents a block diagram of the proposed monitoring and control system in its mode of simplified embodiment, in which the stage of process supervision (13), keeping the rest of the elements of the control system described above, such that the only variation in the operation of the monitoring system and control is that the tracking error (4) is calculated directly based on the temperature measurement provided by the sensor a once filtered and conditioned (12), without any supervision, consequently suppressing time detection capability real malfunction of the monitoring system and control.

In a more detailed way, it indicates then the meaning of the different symbols employees:

one -

Temperature reference setpoint, specify the desired value for the maximum surface temperature of the material during the process.

2 -

Maximum temperature signal on the surface of the material, already validated by the supervision stage (13).

3 -

Adder.

4 -

Tracking error signal, calculated as the difference between the maximum temperature measurement in the surface of the material already validated (2) and the desired value, specified by the reference setpoint (1).

5 -

PID linear controller.

6 -

PID controller actuation signal.

7 -

Neural controller using model reference.

8 -

Signal of the neuronal controller through reference model on the power output of the source To be.

9 -

Material treatment system with To be.

10 -

Process output signal, maximum temperature in The surface of the material.

eleven -

Data acquisition stage, including the infrared pyrometer temperature sensor as well the elements necessary for filtering and signal conditioning measure.

12 -

Maximum temperature signal on the surface of the material measured by the infrared pyrometer, already filtered and conditioned.

13 -

Process supervision stage, verify the coherence between the temperature signal measured by the pyrometer infrared, once filtered and conditioned (12), with the estimated temperature (28) using an incremental neuronal model of process.

14 -

Proportional Gain

fifteen -

Proportional action, action proportional to the value of the tracking error (4).

16 -

Comprehensive gain

17 -

Integrator.

18 -

Integral action, action proportional to the Temporary integral of the tracking error.

19 -

Pure delay.

twenty -

Signal of the tracking error at the instant of previous sampling.

twenty-one -

Adder.

22 -

Temporary derivative of the tracking error, calculated directly from the error difference of monitoring between two consecutive sampling moments (4 and twenty).

2. 3 -

Differential gain.

24 -

Derivative action, action proportional to the temporary derivative of the tracking error (22).

25 -

Adder.

26 -

Anti-saturation mechanism

28 -

Temperature signal estimated by the model in the current instant

29 -

Multilayer perceptron neuronal network.

30 -

Temperature increase in the next instant Of sampling.

31 -

Adder.

32 -

Temperature signal estimated in the following sampling time

33 -

Pure delay.

3. 4 -

Integral temperature increase

35 -

Room temperature.

36 -

Adder.

37 -

Adder.

38 -

Absolute value.

39 -

Divider.

40 -

Difference between the maximum temperature signal measured by the infrared pyrometer, already filtered and conditioned (12) and the temperature estimated by the incremental neuronal model of the process (28), expressed in relative value.

41 -

Acceptance or rejection threshold, expressed in %.

42 -

Comparator If the difference does not exceed the threshold, the supervision stage validates the maximum temperature value in the surface of the material measured by the infrared pyrometer, since filtered and conditioned (12), otherwise it emits an alarm allowing the system operator to take corrective measures timely.

Detailed description of the invention

The present invention introduces a system of monitoring and control of heat treatment processes surface of laser materials by neural control adaptive through reference model, which uses as input variable the maximum temperature on the surface of the material measured by an infrared pyrometer, conveniently calibrated and positioned and as a performance variable the power of Laser source output.

The use of source power output laser as an action variable allows to compensate any deviation detected in the temperature value reached without alter the spatial profile of the temperature distribution in the treated area, thus guaranteeing the uniformity and quality of the treatment.

In this patent two ways of proposing realization of the monitoring and control system: a mode of complete realization in which in addition to monitoring and controlling the process, its development is monitored in real time, checking the consistency between the temperature measured by the infrared pyrometer with the estimated temperature using a model incremental neuronal process incorporated into the system monitoring and control and a simplified embodiment in which suppresses process supervision.

According to the scheme proposed in the figure 1, which corresponds to the complete realization of the system, the monitoring and control system cyclically calculates the signal of performance on the power of the laser source (8) based on the tracking error signal (4), defined as the difference between the maximum desired temperature value, specified by the reference setpoint (1) and the value actually reached by the treatment (2), which has been previously filtered and conditioning (12) and its validity verified by the stage of process supervision (13).

In the event that the maximum temperature signal on the surface of the material already filtered and conditioned (12) not is consistent with the estimated temperature value using the incremental neuronal model of the process (28), the stage of process supervision (13) emits an alarm signal, allowing to the system operator to take corrective measures timely.

The coherence between both signals is verified directly comparing the difference in relative value (40) between the value of the maximum temperature on the surface of the material already filtered and conditioned (12) and the estimated temperature value through the incremental neuronal model of the process (28), with a Acceptance or rejection threshold (41) defined based on the particular characteristics of heat treatment processes of materials with laser and the type of temperature sensor employee.

Essentially, the neuronal controller adaptive by reference model (7), base of the system monitoring and control, allows to assimilate the dynamic response of the surface heat treatment process of materials with laser, highly non-linear and complex response, to the response in open loop of a simple linear performance system known, system called reference model, in this case a first order linear system whose dynamic behavior It is approximately three times faster than the process original.

As a consequence, the set formed by the process and the neuronal controller using reference model is behaves like a simple linear system known first order, system that is easily controllable by a controller Proportional, Integral and Derivative linear error of PID tracking.

From a conceptual point of view, choosing as a reference model the desired behavior for the process in closed loop, the PID linear controller could be dispensed with, but nevertheless, the insertion of a linear type control block PID closing the control loop allows an improvement considerable in the benefits of the whole, mainly in what Regarding its stability and robustness against disturbances External not provided.

In addition, the division of the control system into Two stages, greatly simplify the design and tuning of the operating parameters of the control system, because it is possible to synthesize each stage independently.

In figure 3, it can be seen as by a complex multilayer perceptron neuronal network (27), the Neural controller using reference model (7) calculates the acting on the output power of the laser source (8), in function of the PID controller actuation signal (6) and the status  current process, represented by the maximum temperature value on the surface of the material (2).

Given that the fundamental objective of a controller by reference model is to correct the dynamic behavior of the process, assimilating it to the model of reference, the benefits of the process once controlled will come defined largely by the correct choice of dynamic characteristics of the reference model.

In the patent control system, have established as basic criteria for the choice of the model of reference limit the over-oscillation of the process, that is the maximum deviation of the response with respect to the desired value and also reduce the system establishment time, that is the  time needed for the system to reach status stationary, thereby significantly improving the process efficiency As a result, it has been estimated it is convenient to use a linear system of reference as a reference model first order whose dynamics is approximately three times faster than the original process to control.

Considering the particularities of the surface heat treatment processes of materials with lasers, characterized by a high degree of nonlinearity and a very fast response, the controller using model reference has been implemented using a neural network type multilayer perceptron (27).

Historically, neural networks artificial emerge in an attempt to emulate the mechanisms Basic biological processing and storage, through a complex network consisting of multiple simple units of Parallel processing called artificial neurons, interconnected with each other through a dense connection structure  unidirectional, forming various topologies or architectures in function of the application for which they have been designed.

By their nature, neural networks artificial constitute one of the fundamental techniques for the modeling and control of highly complex nonlinear processes, such as surface heat treatment of materials with laser, for its excellent ability to learn and correct your behavior from a set sufficiently representative of input-output data of process.

According to the type of signals they process, the different neurons in the network can be grouped into homogeneous structures called layers, being possible to distinguish in any typical neuronal network at least three types of layers: the input layer, which groups all the neurons that receive information from the outside, a set of hidden or internal layers that lack contact with the outside and the exit layer, which provides the response of the
net.

They are currently documented in the literature at least about fifty different network architectures artificial neural, being one of the most used in the modeling and control of nonlinear processes known architecture as a multilayer perceptron, a multilayer network with a single layer internal or hidden, in which all the neurons that constitute the network are of the simple perceptron type.

As an artificial neuron, the simple perceptron tries to emulate the operation of the processing units of the sensory systems of vertebrates, being their response a nonlinear function of the neuron input values, function known as activation function or function of transfer.

In accordance with the above considerations, the controller through reference model (7) has been implemented through a multilayer perceptron neuronal network (27), with a inner layer formed by six neurons with activation function of sigmoidal type, that is, a continuous and derivable function of the family of hyperbolic trigonometric functions, in this case concrete, the hyperbolic tangent.

In its complete implementation, the system for monitoring and controlling surface heat treatment processes of laser materials proposed incorporates a process supervision stage (13), which verifies the coherence between the maximum temperature signal on the surface of the material and filtered and conditioned (12) with the estimated temperature value through the incremental neuronal model of the process (28), allowing real-time detection of failures in the operation of the monitoring and control system and taking corrective measures
timely.

As a consequence of the high complexity and not linearity of heat treatment processes of materials with laser, as well as its low response time, for the simulation of Real-time treatment is necessary to resort to techniques of state-of-the-art nonlinear modeling, such as modeling using artificial neural networks to obtain a model of adequate treatment and fast enough for evaluation Direct within the monitoring and control system.

Based on the above considerations, for the modeling of the surface heat treatment process of laser materials a neural model has been developed incremental process, implemented through a neural network multilayer perceptron type, with an inner layer formed by six neurons with sigmoidal activation function.

In the simplified embodiment of the scheme of proposed monitoring and control, according to the structure represented in figure 6, the supervision stage of the process (13), keeping the rest of the elements described above, so that the only variation in its operation is that the tracking error (4) is calculated directly based on the measurement provided by the sensor temperature once filtered and conditioned (12), without any type of supervision, thereby suppressing the ability to real-time detection of system operation failures of monitoring and control.

       \ newpage
    

Embodiment

The proposed monitoring and control system can be carried out by means of a heat treatment equipment surface of laser materials equivalent to the following:

-

A material treatment system with laser (9), which includes a laser source suitable for the type of heat treatment to be performed and that allows real-time control of the output power, in addition to the positioning systems of the beam and the element a treat and other auxiliary elements necessary for its correct functioning. Taking into account the characteristics of the process, it is possible to use a wide variety of laser sources (CO2), Nd: YAG, Diode, etc.), without this influencing the applicability of the invention object of patent.

-

A infrared pyrometer, which is used as a sensor for maximum temperature monitoring on the surface of the material during treatment. Taking into account the characteristics of the process, it is possible to use both pyrometers One color as two colors.

-

A personal computer for process control, provided with card data acquisition, graphic display and other peripherals of usual input / output of data, in addition to software monitoring and control of material treatment processes with laser through adaptive neuronal control using reference.

The connection between the computer and the pyrometer is done through a channel of analog input of the data acquisition card, incorporating a stage of adaptation and filtering of the appropriate signal to the working range of both elements.

In a manner similar, for the control signal of the control system on the power of the laser source is used an analog output channel of the data acquisition card.

In this patent two ways of proposing embodiment: a complete embodiment in which in addition to monitor and control the process, the operation of the same, verifying the coherence between the temperature measured by the infrared pyrometer used as sensor, with the estimated temperature using a neuronal model of the process incorporated into the monitoring and control system and a mode of simplified embodiment in which the stage of process supervision

According to the scheme proposed in the figure 1, in its full realization, the monitoring system and control cyclically calculates the action signal on the power of the laser source (8) based on the error signal of tracking (4), defined as the difference between the value of maximum desired temperature, specified by the setpoint of reference (1) and the value really achieved by the treatment (2), which has been previously filtered and conditioned (12) and its validity verified by the process supervision stage (13).

In the event that the maximum temperature signal on the surface of the material already filtered and conditioned (12) not is consistent with the estimated temperature value using the incremental neuronal model of the process (28), the stage of process supervision (13) emits an alarm signal, allowing to the system operator to take corrective measures timely.

In its simplified implementation, according to the scheme of figure 6, the supervision stage of the process (13), keeping the rest of the elements described above, so that the only variation in its operation is that the tracking error (4) is calculated directly based on the measurement provided by the sensor temperature once filtered and conditioned (12), without any type of supervision, thereby suppressing the ability to real-time detection of system operation failures of monitoring and control.

Industrial application

In general, the system of monitoring and control described is of immediate application in any of the surface heat treatments of materials with lasers usually made in industry, (tempering, normalized, refusion, surface coating, alloy superficial, etc.), only small adjustments being necessary depending on the specific characteristics of the installation and The process to control.

Claims (2)

1. Monitoring and control system surface heat treatment processes of laser materials by adaptive neuronal control by reference model comprising the following elements:

-

Material treatment system with laser, which includes a laser source, positioning equipment of the beam and the element to be treated and other auxiliary elements necessary for proper operation (7).

-

Infrared pyrometer used as maximum temperature sensor on the surface of the material.

-

Personal computer for the control of process, provided with data acquisition card, screen graphics and other peripherals for data input / output, in addition to a software for monitoring and control of treatment processes of materials with laser by adaptive neuronal control by reference model.

characterized in that:

-

Cyclically, the maximum temperature in the surface of the material to be treated (12), measured by the infrared and conveniently filtered and conditioned pyrometer, is compared to the desired process temperature value, specified by the reference setpoint signal (1), entered through the data entry device of the computer that controls the process, getting the error of tracking (4), which is processed by the system monitoring and control, generating the signal of action on the output power of the laser source (8),

-

He control system is divided into two stages: a controller Linear Proportional, Integral and Derivative type, PID (5), which generates the corresponding actuation signal (6) based on the error tracking (4) and secondly a neuronal controller adaptive by reference model (7), implemented by a multilayer perceptron neuronal network, which calculates the signal of action on the output power of the laser source (8) a from the actuator signal of the PID controller (6) and the maximum material surface temperature (12), measured by the infrared pyrometer, filtered and conditioned, and

-

He PID linear controller (5) calculates the actuation signal corresponding (6) based on the tracking error (4), as a linear combination of three basic control actions: one action proportional to the follow-up error (15), an action proportional to the temporary integral of the tracking error (18) and an action proportional to the temporary derivative of the error of follow-up (24).

2. System for monitoring and controlling surface heat treatment processes of laser materials, according to claim 1, characterized in that it incorporates a process supervision stage (13), which verifies the coherence between the maximum temperature signal on the surface of the material measured by the infrared pyrometer, once filtered and conditioned (12), with the estimated value of maximum temperature on the surface of the material (28), which is estimated by a neuronal model of the process implemented incrementally by a completely neural network independent (29), multilayer perceptron type with six neurons in its inner layer, which estimates in real time the increase in the maximum temperature on the surface of the material at the next instant of sampling (30) based on the maximum temperature in the surface of the material estimated by the model at the current time (28) and the performance on the power output of the source laser (8) at the present time, and in the event that the maximum temperature signal on the surface of the material measured by the temperature sensing pyrometer once filtered and conditioned (12) is not consistent with the temperature value estimated by the neuronal model of the process (28), the process supervision stage (13) emits an alarm signal, allowing the appropriate corrective measures to be taken.