System and procedure for controlling an actuator

Abstract

Comprehensive actuator-sensor fusion system, at least one real sensor (108,212) connected to an actuator (104,205) and generating real sensor signals, generating at least one virtual sensor (110) virtual sensor signals that define an estimated position of the actuator, based on signals used to control the actuator (104, 205), and a processor for merging the real sensor signals and the virtual sensor signals in order to detect a failure of the actuator (104,205) and / or of at least one real sensor (108,212), wherein the processor is configured to collect periodic corrections and / or absolute values ​​of the periodic corrections applied to the virtual sensor (110) and to compare the collected periodic corrections or absolute values ​​of the periodic corrections with a limit in order to detect deterioration of the actuator (104,205) and / or the real sensor (108,212), and a memory factor (k ff ) at predetermined time durations (t ff ) to apply to the accumulating periodic corrections or absolute values ​​of the periodic corrections, so that the accumulating periodic corrections or absolute values ​​of the periodic corrections are periodically reduced by a certain proportion of their current values.

Description

[0001] The present invention relates generally to actuators used in control processes in production plants. It relates in particular to a system and a method for controlling an actuator, wherein the positioning of actuators and the validation of the actuator's position sensor are controlled by combining or merging data from physical or real sensors with data from virtual sensors, both of which are used to monitor the actuators. While it is obvious that the approach of the present invention can be used in a wide variety of applications, the invention is described here with reference to actuators used in papermaking machines, for which the invention was developed and initially used.

[0002] Systems with feedback control are known from the prior art as a means of controlling process variables in production facilities, for example, for positioning an actuator in a desired position. Feedback mechanisms and control algorithms are used to reduce a detected fault signal to a minimum value. However, the prerequisite for control is that the feedback position is that of a fault-free state. DE 101 37 597 A1 describes a method for fault diagnosis on a clutch actuator.

[0003] If sensors are used to monitor actuators, they are typically single-element feedback devices. Such sensors are capable of providing good feedback under either dynamic or near-static conditions, but usually not both. The level of precision and accuracy required for such sensors, especially those suitable for broadband operation, often exceeds what is commercially available at a practical cost for use in industrial control applications.

[0004] If actuator sensors suddenly fail, the failures are generally simply detected, with the unfortunate result that the feedback from the failed sensor is completely unusable. Under these circumstances, the only course of action is to avoid any further action to control the actuator in order to prevent potentially catastrophic or permanent damage to system components, which would severely disrupt the operation of the machine in which the actuators are used.

[0005] If the sensor's noise level is close to the signal level required for control, closed-loop control systems can introduce fluctuations into the controlled process. This phenomenon has been observed in papermaking machines, particularly with sensitive headbox control systems. The current solution to such fluctuation problems is to widen the dead zone, which reduces control accuracy and leads to shorter process response times.

[0006] It has also been observed that very slow-developing drifts or errors, such as those emitted by progressively deteriorating sensor hardware, can be masked within the closed control loop, a typical characteristic of actuators with control systems. Such masked drifts or errors can render the control mechanism and / or typical alarm structures completely useless.

[0007] A system and method for improved actuator control generates hybrid, soft, or smart sensors by merging information generated by at least one physical or real sensor with information generated by at least one virtual sensor. Virtual sampling can be performed by accumulating control signals and / or the absolute values ​​of the control signals applied to an actuator, for example, by counting steps applied to a stepper motor, to effectively integrate the control signals and / or their absolute values. The accumulated control signals are then transformed into a corresponding position where the actuator would be based on the accumulated control signals.The resulting virtual position of the actuator, estimated as accurately as possible, is used together with the physically detected position of the actuator, also estimated as accurately as possible, by the real sensor, to form the hybrid, soft, or smart sensor. In this context, the terms "fused," "merging," or "fusion" refer to the use or combination of the signals to form the hybrid sensor.

[0008] The accumulating control signals can be subjected to minor periodic corrections to maintain or achieve a correlation between the virtual sensor and the real sensor over time, if possible. These corrections gradually reduce errors in the actuator position indicated by the virtual sensor. To mitigate accumulating numerical errors in the accumulating control signals and also to reduce the significance of events that occurred long ago, a displacement factor (k) can be applied. ff ) at fixed time durations (t ff ) are used so that the accumulating control signals are reduced by a certain proportion of their current values.

[0009] According to one aspect of the teachings of the present application, a system for merging a sensor for an actuator comprises at least one real sensor connected to an actuator and generating real sensor signals, and at least one virtual sensor generating virtual sensor signals based on signals used to control the actuator. A processor merges the real sensor signals and the virtual sensor signals to detect a malfunction of the actuator and / or the at least one real sensor. The processor can be configured to revert to operating the actuator using the at least one virtual sensor when a malfunction of the at least one real sensor is detected. The processor can model the at least one virtual sensor and collect historically recorded data representing the control signals sent to the actuator.

[0010] The processor can be configured to combine the signals of the real sensor and the signals of the virtual sensor by comparing a position of the actuator indicated by the signals of the real sensor with a position of the actuator indicated by the signals of the virtual sensor, in order to determine a difference in the indicated positions of the actuator and to determine a malfunction of the actuator and / or of at least one real sensor due to the difference in the indicated position exceeding a difference limit.

[0011] The processor can be configured to accumulate periodic corrections made with respect to the virtual sensor and to compare these accumulated corrections to a limit in order to determine any degradation of the actuator and / or the real sensor. The processor can also be configured to incorporate a memory factor (k). ff) at certain time durations (t ff ) to apply to the accumulating periodic corrections, so that the accumulating periodic corrections are periodically reduced by a certain proportion of their current values.

[0012] The processor can also be configured to accumulate absolute values ​​of the periodic corrections to which the virtual sensor is subjected and to compare the accumulated absolute values ​​of the periodic corrections with a limit in order to detect deterioration of the actuator and / or the real sensor. The processor can further be configured to implement a memory factor (k ff ) at fixed time periods (t ff ) to apply to the accumulating periodic corrections and a memory factor (k ff1 ) at fixed time durations (t ff1) to apply to the accumulating absolute values ​​of the periodic corrections, so that the accumulating periodic corrections and the accumulating absolute values ​​of the periodic corrections are periodically reduced by certain ranges of their current values.

[0013] The processor can be trained to accumulate the absolute values ​​of the periodic corrections to which the virtual sensor is subjected and to compare the accumulated absolute values ​​of the periodic corrections against a limit in order to detect degradation of the actuator and / or the real sensor. In this case, the processor can be trained to implement a memory factor (k ff1 ) at fixed time durations (t ff1 ) to apply, so that the accumulating absolute values ​​of the periodic corrections are periodically reduced by a certain proportion of their current values.

[0014] According to another aspect of the teachings of the present application, a method for merging a sensor for an actuator using a processor comprises monitoring at least one real sensor connected to an actuator, monitoring at least one virtual sensor using a processor, and merging real sensor signals generated by the real sensor and virtual sensor signals generated by the virtual sensor using a processor in order to detect a malfunction of the actuator and / or the at least one real sensor. The method may also include, based on a detected failure of the at least one real sensor, reverting the actuator to operation using the at least one virtual sensor.

[0015] The method can further include modeling the at least one virtual sensor and accumulating historical tracking data representative of control signals sent to the actuator. Merging the signals of the real sensor and the virtual sensor can include comparing the actuator's position indicated by the real sensor's signals with the position indicated by the virtual sensor's signals, determining any difference in the actuator's displayed position, and indicating a malfunction of the actuator and / or the at least one real sensor based on a difference in the displayed position exceeding a specified limit.

[0016] The method may further include accumulating periodic corrections applied to the virtual sensor and comparing the accumulated periodic corrections with a limit to determine degradation of the actuator and / or the at least one real sensor. In this case, the method may also include applying a memory factor (k ff ) at predetermined time intervals (t ff ) include the accumulating periodic corrections, so that the accumulating periodic corrections are periodically reduced by a certain proportion of their current values.

[0017] The method may further include accumulating both periodic corrections applied to the virtual sensor and the absolute magnitude of periodic corrections applied to the virtual sensor, and comparing the accumulations with limits to detect degradation of the actuator and / or the at least one real sensor. In this case, the method may further include applying a memory factor (k) using a processor. ff ) at fixed time intervals (t ff ) on the accumulating periodic corrections and using a processor, applying a memory factor (k ff1 ) on the accumulating absolute values ​​of the periodic corrections at fixed time intervals (t ff1) include such that the accumulating periodic corrections and the accumulating absolute values ​​of the periodic corrections are periodically reduced by a certain proportion of their present values.

[0018] The method may further include, using a processor, accumulating the absolute values ​​of periodic corrections applied to the virtual sensor and comparing the accumulated absolute values ​​of the periodic corrections with a limit to determine degradation of the actuator and / or the real sensor. In this case, the method may further include, using a processor, applying a memory factor (k). ff1 ) at fixed time intervals (t ff1 ) include, so that the accumulating absolute values ​​of the periodic corrections are periodically reduced by a certain proportion of their current values. Brief description of the drawings:

[0019] The features and advantages of the invention according to the present application will be apparent to those skilled in the field to which the invention relates from the following description of the embodiments shown and the attached claims together with the accompanying drawings, in which: Fig. 1. A block diagram of a system operable according to the teachings of the attached application shows, Fig. 2 a block diagram of an exemplary embodiment of a hybrid sensor according to the teachings of the present application shows, Fig. 3. Processes of an exemplary embodiment of a hybrid sensor according to the teachings of the present application, Fig. 4. Processes of an exemplary embodiment of a hybrid sensor according to the teachings of the present application for a slow drift of a hardware sensor are presented, Fig. 5. Procedures of an exemplary embodiment of a hybrid sensor according to the teachings of the present application for a slight but permanent engine creep, Fig. 6. Procedures of an exemplary embodiment of a hybrid sensor according to the teachings of the present application for a sudden failure of a hardware sensor, and Fig. Figure 7 shows the processes of an exemplary embodiment of a hybrid sensor according to the teachings of the present application for a continuous motor. Detailed description of the invention:

[0020] The system and method of the present application are described with reference to the control of actuators in paper-making machines, for which it was developed and initially used. However, those skilled in the art will recognize that the present invention can be used in a wide variety of applications.

[0021] A system and method for improved actuator control are disclosed to create hybrid sensors, also referred to herein as soft sensors or smart sensors, by merging information generated by at least one real sensor with information generated by at least one virtual sensor. Papermaking machines employ numerous actuators to control the process. For example, numerous headbox actuators are used to control the edge opening of a headbox outlet. These headbox actuators require a high degree of accuracy to precisely position the outlet edge.Most discharge edge actuations are small and occur gradually over time, so if actuator sensors are used, they must function well with small, slowly varying movements that exhibit only minor changes. However, some discharge edge actuations are relatively large and require rapidly changing movements. Relatively large, rapidly changing movements can occur, for example, globally across the entire headbox when paper grade changes are performed. The larger actuator movements also require a high degree of accuracy to ensure that safety measures against edge bending are maintained, preventing damage to the discharge edge during these large movements.Therefore, actuator sensors used to monitor actuators in headboxes must also function reliably during relatively large and rapid actuator movements. Consequently, actuator sensors are required that function well under both relatively static and relatively dynamic conditions.

[0022] Since wide-bandwidth sensors are typically expensive, their use may not be practical in many applications, particularly those requiring a relatively large number of actuators, such as headbox edge control, where many actuators and correspondingly many sensors are needed for the wide width of papermaking machines. In accordance with the teachings of the present application, information from various sources is intelligently merged to extend the functionality of sensors beyond the usual simple comparison of their output signals to the limits of those signals, thus enabling the use of commercially available, relatively inexpensive sensors.

[0023] Knowing the properties of an actuator allows for the development of an open-loop control model and its tracking history to create a virtual sensor. This virtual sensor's tracking history is then used to generate secondary measurements (MVs), or estimates of the actuator's movements or position. These estimates are compared to measurements taken from a physical sensor monitoring the actuator. Therefore, methods are employed to merge data from primary measurements generated by a physical sensor with those generated by a virtual sensor. This allows for higher-quality measurements and greater measurement consistency through error estimation, tracking, and system adjustment.

[0024] Fig.Figure 1 shows a simplified system 100 that can be operated according to the principles of this application. A process controller 102 controls an actuator 104, typically one of numerous actuators, via an actuator control unit 106. A hardware sensor 108 monitors the actuator 104 and generates physical sensor signals or real sensor signals that are representative of the measured actuator movement. The sensor 108 can take any form suitable for a given application, such as a conventional linear variable differential transformer (LVDT) for linear actuator applications, a rate gyroscope for rotary actuators, and even rate gyroscopes at a suitable location in linear actuator units with appropriate modeling. Using the specifications of the sensor 108 and normal statistical and / or systemic error characteristics, as well as measurements performed on the sensor 108, the operating characteristics of the sensor 108 are determined.The determined operating characteristics are then used to generate an open control model designed for normal operating conditions of the actuator / sensor combination 104 / 108. Control modeling is well known to those skilled in the art and is not described beyond what is necessary for an understanding of the system and the method for improved actuator control according to the teachings of this application.

[0025] A virtual sensor 110 continuously tracks issued instructions (or a stepping motion if applied to a stepper motor) to generate virtual sensor signals that define an estimated position of the actuator 104. The output signals or estimated signals of the virtual sensor 110 for the actuator's position tend to be very accurate in short-term and step-by-step modes, but tend to drift slowly out of accuracy because the actuator is a physical device and does not respond in exact accordance with the instructions sent for its control. For example, steps may be executed incorrectly due to low torque during the early steps of a stepper motor. Because the virtual sensor 110 is an integrating device, the sensor drift error can become infinitely large over a long period of time.

[0026] Signals from the physical sensor or hardware sensor 108 are compared with signals from the virtual sensor in a sensor fusion device 112 to detect and track differences between the physically determined position of the actuator 104 and the calculated position of the actuator 104 determined by the virtual sensor 110. The data from the real sensor and the data from the virtual sensor are combined to achieve a higher quality of position data through the hybrid sensor, soft sensor, or smart sensor, as defined by the teachings of this application. The sensor fusion is not limited to one physical and one virtual sensor; rather, in accordance with the teachings of this application, numerous data sources can be fused.The virtual sensor 110 and the sensor fusor 112 can be integrated into one or more associated processors or into a processor connected to the control of the system using the actuator 104, including the virtual sensor 110 and the sensor fusor 112. The hybrid, soft, or smart sensor can also, in a desirable manner, determine more information regarding the integrity of the sensor 108 and the actuator 104, thereby increasing the intelligent utility of the equipment.

[0027] Fig.Figure 2 is a block diagram of a possible embodiment of a hybrid sensor 200 in accordance with the teachings of the present application, wherein the actuator is represented as a stepper motor. Control setpoint changes are directed to a stepper generator 202, which converts the setpoint changes into a suitable number of steps, which are then sent to a stepper motor 204, an actuator 205. The step signals are directed to a time-delay compensator 206 and a step counter 208, which counts the steps in order to integrate the control signals directed to the actuator 205 through the compensator 206 as part of the virtual sensor. The time-delay compensator 206 adjusts the number of steps to account for a delay in a gearbox 210 of the actuator 205 when changes of direction are performed.For example, if a reverse movement of ten steps is required, fifteen steps may be necessary to compensate for the overrun in the connected actuator gear 210. The operating characteristics of the overrun compensator 206 are determined empirically in the usual manner. A hardware sensor 212 is connected to the actuator 205 in the usual manner to monitor the position of the actuator, as is known from the prior art.

[0028] While the output signal of sensor 212 can be used directly, in the Fig.In the embodiment shown in Figure 2, the sensor's output signal is modified to account for the physical properties of the sensor 212 and the system and environment in which the sensor is used. For example, if the sensor 212 is moved in one direction and then back in the opposite direction, there is a slight mismatch or hysteresis in the sensor 212's output signal, similar to the overrun in the gearbox 210 of the actuator 205. This mismatch is compensated for by using a hysteresis corrector 214. The hysteresis of the sensor 212 can be determined based on the sensor 212's specifications and / or by tests performed on the sensor currently in use. Furthermore, a variability corrector 216 is provided to compensate for deviations from the ideal properties of the sensor 212, since the sensor 212 is a real, physical device that deviates from nominal characteristics.For example, a linear sensor with linear properties will typically deviate from an ideal degree, so that if nonlinear corrections are made, these will improve the accuracy of the sensor's output signal. The sensor 212 can also be adjusted to accommodate other variations from the ideal state, indicated by a tuning corrector 218. The tuning corrector 218 can, for example, compensate for sensor variations due to temperature.

[0029] The direct output signal (or the compensated / adjusted output) of sensor 212, which represents the best possible assumption of the position of actuator 205 based on measurements taken by sensor 212, is then routed to a diagnostic / analysis module 212 for merging with a best possible assumption of the actuator's position generated by the virtual sensor. The virtual acquisition is performed according to the embodiment shown. Fig.2 is carried out by the pedometer 208, which counts the steps applied to the actuator 205 by the step generator 202. The pedometer 208 effectively integrates the steps received from the step generator 202. The output signal of the pedometer 208 is converted into a corresponding position in which the actuator 205 would ideally have moved if the actuator 205 had received the number of steps collected by the pedometer 208 from a step converter 222. The calculated resulting best-estimated position of the actuator 205, which is the output signal of the virtual sensor of the embodiments according to Fig. 2 is also directed to the diagnostic / analysis module 220 to be merged with the best possible estimated position of the actuator 205 by the sensor 212.

[0030] As will be explained in more detail below, the pedometer 208 is subjected to minor periodic corrections by means of a periodic corrector 224 if limiting circumstances exist, in order to maintain or achieve a correlation between the virtual sensor and the hardware sensor over time, if possible. The corrections made by means of the periodic corrector 224 gradually reduce numerical errors regarding the position of the actuator as indicated by the virtual sensor.These numerical errors can be caused by rounding processes resulting from the finite accuracy of real floating values ​​in numerical calculations, the discrete nature of steps, sometimes missed motor steps during start-up operations, which may result from unenergized step coils or step coils with a low holding current that are not in use to reduce quiescent current requirements, misalignment of the home position, possible sensor drift, or the like. In working embodiments of the system and method for improved actuator control according to the present application, periodic corrections were performed once every minute at a size of 0.7 micrometers (for linear corrections) or 0.1 degrees (for rotary corrections).

[0031] With this understanding of the system and method for improved actuator control according to the teachings of the present application, statistical tracking and sensor merging are now described to provide a better understanding of a presented embodiment of the invention. To track the performance characteristics of a sensor and a motor, the following statistics are recorded: - The variance of the current between the hardware sensor and the virtual sensor (H2SΔ) - The sum of periodic corrections (direction-sensitive) (Σ(correction)) - The sum of absolute periodic corrections (not "direction" sensitive) (Σ|correction|)

[0032] These statistical values ​​are reset to zero when the machine system, including the actuator, is started and are collected over the duration of each actuator's operation. A memory factor (k) can be used to mitigate the effects of accumulating numerical errors in the statistical values, reduce the significance of events that occurred long ago, and thereby give greater weight to recent events. ff ) are applied. By periodically using the memory factor (k ff ) based on a fixed time period (t ff ) the statistical accumulators are reduced by a certain proportion of their current values. For example, (t ff ) is set to one hour, and (k ff ) is set to a value of 0.90, the following applies at hourly intervals: Σ(Correction)=Σ(Correction)•0.90; Σ|Correction|=Σ|Correction|•0.90

[0033] The memory factor also helps to reduce gross random effects that falsely trigger the detection of problems, and in accordance with the teachings of the present application, different memory factors and different time durations can be used for application in the memory factor.

[0034] The values ​​of (t ff ) and (k ff These values ​​are determined based on considerations regarding the desired detection rates for sensor drift, motor creep, or the like. Shorter time periods and / or smaller proportional settings increase the amount of drift, motor creep, or the like required to detect a problem. In one working implementation, (t ff ) set to a value equal to one hour and (k ff) was set to a value of 0.98. By using a memory factor, the possibility of a correction made in the distant past influencing the triggering of any alerts during the current time is reduced.

[0035] If a memory factor is used, a 1% correction that occurred an hour ago and another 1% correction that was needed now should trigger an alarm. However, a 1% correction that occurred seven months ago and the same 1% change that is occurring now are not cause for concern, so no alarm should be triggered. It should be noted that careful tuning of the rate of periodic corrections, the size of the periodic corrections, the alarm averages, and the memory rate is necessary. For example, a memory factor (k ff) of 0.95 and a time factor (t ff ) from two hours, little more than a day, approximately 26 hours to reduce the importance of an event by 50%, another day to reduce the historical importance of an event to 25%, etc.

[0036] Σ(Correction) is an indication of sensor drift, and Σ|Correction| is an indication of motor creep, considering that a motor is usually steered to approximately the same position and creep can occur in both directions, expected to average evenly over time. Although described as statistical values, it is noted that H2SΔ, the Σ(Correction) accumulator, and the Σ|Correction| accumulator are only indicators and not mathematically strong statistical values, since both creep and drift will have stochastic effects on both accumulators.

[0037] Fig.Figure 3 shows comparisons between the values ​​of the (H2SΔ) of the hardware sensor and the virtual sensor, and variance sections in which the periodic corrector 224 is active. The different limit values ​​are shown as 1% (L1) and 2% (L2) of the total sensor range. The difference limit values ​​shown, L1 and L2, were chosen for testing purposes, and it is obvious that, using the teachings of the present application, other percentages, fixed values, or other suitable values ​​can be used for given applications. Fig. Figure 3 shows that if the variable H2SΔ between the hardware sensor and the virtual sensor lies within + / - 1% of the total sensor range of mutual agreement, the virtual sensor is set as good, and the value of the virtual sensor is generally used because it has the higher resolution.

[0038] If the hardware and virtual sensors are substantially in agreement, such that the variance H2SΔ between the hardware and virtual sensors is between 1% and 2% of the total sensor range of mutual agreement, the variance is considered negligible and the periodic corrector 224 is activated. For example, if the virtual sensor reading was increased by 1.5%, perhaps due to poor initialization, a brief, one-time motor stall, uncertainty during startup regarding the direction in which the motor last moved, or similar factors, then after a certain period of time the periodic corrections would gradually bring the hardware and virtual sensors back into an acceptable alignment or correlation, that is, into agreement within 1%.If such a one-off event has been corrected and no further symptoms have occurred, the system has been safely corrected with respect to the sensor discrepancy.

[0039] If, as in the illustrated embodiment of Fig. If value 3 is displayed, and a larger error is detected because the variance H2SΔ between the hardware sensor and the virtual sensor is greater than 2% of the total sensor range (failure range), a decision is made as to which result should be used. In the case of a motor failure, the hardware sensor should be used, as this indicates the physical output. If a hardware sensor failure is detected, the virtual sensor provides an indication of the physical output state.

[0040] The following examples represent responses using the monitored statistical values ​​in accordance with the teachings of the present application. The first, in Fig.Example 4 (where +Det and -Det are positive and negative average detection values, respectively) shows a slow drift of the hardware sensor, a condition that would previously have gone unnoticed by control systems for an actuator. Previously, any small error (below the average value of an unexpected movement) occurring between control scans was subsequently masked by the feedback from the quality control system (QCS) performing a control action. The drifting sensor would ultimately only have been detected if an actuator had exceeded the bending limit of the outlet edge, even though the outlet / actuator would have been physically located where it was required from a process control perspective. This condition could have persisted for several hours, as bending limits are typically ±500 micrometers, requiring a sensor drift of well over 500 micrometers. All statistical values ​​of the Fig. Figure 4 shows a value diverging from the ideal center lines. When detecting the drifting sensor using the teachings of the present application, one possible action would be to trigger an alarm indicating that the sensor is suboptimal. By detecting the drifting sensor early, instead of waiting for an actuator limit to be exceeded, it is not necessary to disable the actuator at that point.

[0041] The second example, which in Fig.Figure 5 shows a slight but persistent motor creep. Because slight motor creep is not completely detrimental to control performance, it can cause the actuator and the system using the actuator to feel sluggish. Although the QCS corrects for creep on successive scans, detecting slight creep is useful to provide an indication of a failing motor, which can be corrected during scheduled preventive maintenance. A motor that begins to show signs of creep can therefore be scheduled for replacement before a complete failure impacts headbox performance by causing an unplanned outage.If motor slippage is detected, possible actions using the teachings of the present application would be to trigger an alarm that the motor is not in optimal condition, to set a flag to increase a torque setting to control the motor, and to indicate that shutting down the motor is not necessary at the present time.

[0042] The third example, which is in Fig.Figure 6 represents a sudden hardware sensor failure. Large sensor shifts characteristic of a hardware failure, a target shift, or the like are detected by large and sudden movements in the H2SΔ statistics. In the case of a failed or malfunctioning sensor, the sensor may hang after the shift, as shown in (a), or continue tracking, as shown in (b). If the shift in the H2SΔ value is large enough, no periodic corrections are applied. When operating the exemplary embodiment of a hybrid sensor according to Fig.3. For example, no periodic corrections are applied for a shift of the H2SΔ value by more than 2% outside the range. As is evident, detection of hardware sensor failures is achieved by comparing the statistical H2SΔ values ​​against limit values. A typical response to an alarm would be an alert signal indicating that an unexpected movement has occurred, suggesting a faulty sensor. An alert signal for a faulty sensor would typically result in the actuator being locked in place by preventing further step requests from being sent to the actuator. However, if a hybrid sensor is used according to the teachings of the present application, the machine using the actuator can continue to operate, at least intermittently, using data generated by the virtual sensor.

[0043] The fourth example, which is in Fig.Figure 7 shows a continuously running motor. In the case of a continuously running motor, the control system has not instructed any steps, so the virtual sensor will not increment. However, the hardware sensor, which measures the physical movement of the motor, detects an increasing H2SΔ value. Periodic corrections of Σ(correction) and Σ|correction|, even if slow and small, will most likely prevent Σ(correction) and Σ|correction| from reaching the detection limits ±Det, thus triggering the H2SΔ limit. The control system should take any available action to stop the continuous running. However, it is likely that a catastrophic failure of the motor drive circuit has already occurred.When detecting motor creep using the teachings of the present application, possible actions would also include switching off the stepper drive lines, stopping the drive, and zeroing the reference circuit of the digital / analog converter (DAC) to limit current.

[0044] Considering the above various examples, the combination of the observation of the three statistical values ​​can be summarized in the following table. H2SΔ limit Σ(Correction)@Limit Σ|Correction |@Limit diagnosis Actions < L1 No No Normal operation normal operation > L1, < L2 > Det or > Det Slow sensor drift Indicates faulty sensor, continue operation Periodic > L1, always < L2 Low value > Det Low engine glide Engine failing to display, continue operation > L2 Low value Low value Sudden sensor displacement or suspended motor indicator failing actuator, motor lock > L2 Increasing value (up to L2) Increasing value (up to L2) Continuous engine indicator failing actuator, motor lock

[0045] It should be noted that while there may be some uncertainty in differentiating a sudden sensor offset from a continuously moving motor element (the statistical accumulators can be difficult to place), a large H2SΔ value in both cases indicates that motor shutdown is required. If the periodic correction accumulators select a fault while the H2SΔ value is below the alarm average, a slower-developing fault type is usually indicated, allowing the system to continue operating; however, components should be replaced at the next opportunity. The H2SΔ statistics are most important for deciding when immediate action is required.

[0046] Although the invention of the present application has been described with particular reference to its specific embodiments, variations and modifications of the present invention can be made within the idea and scope of the following claims. In particular, the hybrid sensors of the present application can combine the best features of each information source in an optimal manner to enable improved tracking of actuator and sensor characteristics for enhanced diagnostics and allow for self-determination of sensor faults as well as continued operation of systems that use actuators monitored by hybrid sensors in emergency operating mode.

[0047] Actuators with or without sensors do not offer the precision required for modern headbox applications. If a sensor failure occurs and is sufficiently abrupt to be detected, it is best to disable the actuator, resulting in significantly reduced headbox performance. The hybrid sensors of this application enable intelligent inline diagnostics of such failures, as they allow the actuator to operate in a fail-safe mode, using the calculated position from the remaining information source(s) for continued operation. Slowly occurring faults / deviations due to progressive sensor failure can be masked by the similar nature of typical control applications and lead to serious damage, such as permanent deformation of the headbox outlet.According to the teachings of the present application, slow sensor deviations and failure modes can be detected through continuous diagnostics due to the continuous cross-referencing between the various information sources.

[0048] The hybrid sensors of this application offer improved broadband response with low noise, precision, accuracy, and stable measurements under both low dynamic process conditions and fast dynamic process conditions during periods of rapid actuator movement. They also enable enhanced diagnostics of actuators and sensors. By cross-referencing or merging sensor information with other information sources, rather than simply comparing sensor information with overall limits, the hybrid sensors of this application allow for continuous actuator operation during periods of partial sensor failure, while providing warnings so that appropriate maintenance can be performed at the next available opportunity, thereby maximizing the uptime of the entire control system.

[0049] By combining feedback information from various locations / sources using intelligent data fusion, it is also possible to detect error types including abrupt failures, high noise, and deviation errors. If desired, in the event of a failure of one of the two data sources, a return to single-sensor operation, as described, is possible. For applications requiring high operational readiness that necessitates continuous operation, such as paper manufacturing machines where machine downtime is extremely costly, the hybrid sensors allow operation to continue in a reduced or emergency mode, which is exceptionally valuable.Furthermore, the intelligent combination of data sources with different dynamic properties enables optimal use of sensor characteristics such as low noise, stable precision in slowly changing environments and fast dynamic responses for rapidly changing process environments.

[0050] The data fusion techniques according to the teachings of the present application can be implemented either with an actuator unit itself or with a higher-level control architecture. The fusion techniques can incorporate information from two or more sources, real and / or virtual, to create an optimal hybrid sensor, smart sensor, or soft sensor.

Claims

[1] Actuator-sensor fusion system comprising, at least one real sensor (108,212) connected to an actuator (104,205) and generating real sensor signals, generating at least one virtual sensor (110) virtual sensor signals that define an estimated position of the actuator, based on signals used to control the actuator (104, 205), and a processor for merging the real sensor signals and the virtual sensor signals in order to detect a failure of the actuator (104,205) and / or of at least one real sensor (108,212), wherein the processor is configured to collect periodic corrections and / or absolute values ​​of the periodic corrections applied to the virtual sensor (110) and to compare the collected periodic corrections or absolute values ​​of the periodic corrections with a limit in order to detect deterioration of the actuator (104,205) and / or the real sensor (108,212), and a memory factor (k ff ) at predetermined time durations (t ff ) to apply to the accumulating periodic corrections or absolute values ​​of the periodic corrections, so that the accumulating periodic corrections or absolute values ​​of the periodic corrections are periodically reduced by a certain proportion of their current values. [2] System according to claim 1, characterized by, that the processor is designed to revert to operation of the actuator (104,205) using the at least one virtual sensor (110) after a detected failure of the at least one real sensor (108,212). [3] System according to claim 1 or 2, characterized by , that the processor models at least one virtual sensor and collects historical tracking data representative of control signals sent to the actuator (104,205). [4] System according to any one of the preceding claims, characterized by, that the processor is designed to fuse the real sensor signals and the virtual sensor signals by comparing a position of the actuator (104,205) indicated by the real sensor signals and a position of the actuator (104,205) indicated by the virtual sensor signals in order to detect a difference in the displayed position of the actuator (104,205) and to indicate a failure of the actuator (104,205) and / or of the at least one real sensor (108,212) due to exceeding a difference limit by the difference in the displayed position. [5] Actuator-sensor fusion method, comprising: a monitoring of at least one real sensor (108,212) connected to an actuator (104,205) using a processor to generate real sensor signals that define a detected position of the actuator, monitoring at least one virtual sensor (110) using a processor to generate virtual sensor signals that define an estimated position of the actuator, and a fusion of the real sensor signals and the virtual sensor signals using a processor to detect a failure of the actuator (104,205) and / or of at least one real sensor (108,212), a collection of periodic corrections applied to the virtual sensor (110) and a comparison of the collected periodic corrections and / or the absolute values ​​of the periodic corrections with a limit in order to detect a deterioration of the actuator (104,205) and / or of the at least one real sensor (108,212), and applying a memory factor (k ff ) at predetermined time durations (t ff) on the accumulating periodic corrections or absolute values ​​of the periodic corrections, so that the accumulating periodic corrections or absolute values ​​of the periodic corrections are periodically reduced by a certain proportion of their current values. [6] Method according to claim 5, further comprising a return to the operation of the actuator (104,205) using the at least one virtual sensor (110) after a detected failure of the at least one real sensor (108,212). [7] Method according to claim 5 or 6, further comprising modeling the at least one virtual sensor and collecting historical tracking data representative of control signals sent to the actuator (104,205). [8] Method according to any one of claims 5 to 7, comprising fusing the real sensor signals and the virtual sensor signals: a comparison of an actuator position (104,205) indicated by the real sensor signals and an actuator position indicated by the virtual sensor signals, determining a difference in the displayed position of the actuator (104,205) and an indication of a failure of the actuator (104,205) and / or of at least one real sensor (108,212) due to the difference in the displayed position exceeding a difference limit.

Images