CA2685575A1 - Vital wayside train detection system - Google Patents

Abstract

An track-mounted train detection system is disclosed. The invention describes the sensor detection apparatus and methods based on the principle of diversity. The intelligent wheel flange sensor system uses sensors on both rails separated by an optimum distance so as to contain an entire bogie therein. This allows classification of cars, detection of high rail vehicles, maintenance vehicles and locomotives by virtue of the different wheel sequences as the vehicle moves over the sensors. This invention uses time, space and wheel diversity in measurement of each axle and produces track occupied, direction, train speed, train length, wheel count, length of train, end of train detection and computes train progression by integration of the speed-time curve. The sensor also detects wheel shock and vibration for alarming of high wheel impacts.
The sensors incorporates real-time static and between-wheel calibration, self monitoring and unique processing algorithms to prove the sensors ability to function safely in a vital application.
A self test capability tests the sensors dynamically through a programmable self test magnetic stimuli that emulates a complete train movement at any speed over the sensors, such that all system responses to said stimulus is identical to that of a real train. The sensor system is used to create virtual blocks of any length which coupled with the train progression data, create both variable moving block and fixed block train control systems.

Description

Background Most inductor-based train wheel sensors in use today make use of the eddy currents induced into the train wheel to create measurable voltage changes across the inductor of the resonating drive sensor coil. The eddy currents create losses by reducing the flux which results in a change in the voltage of the coil and results in an increase in drive oscillator resonating frequency and a lower coil voltage. Traditional wheel sensors and axle counters are subject to many wheel detection failure mechanisms such as electrical interference, noise, marginal flange height and positioning, sensor movement and rail vibration etc, which renders these systems impractical for vital train detection. Furthermore, most sensor systems do not have the necessary self monitoring and internal testing facilities integrated within to detect or mitigate these failures.

Wheel or Axle counting systems count the wheels and the accuracy of the count is essential. In virtual block systems, the block occupancy is determined by counting wheels in and wheels out of the block. A missed wheel detection is problematic since it is not known if the train is still in the block or not. Being failsafe, we treat the missed wheel as if the block was still occupied and must be reset after the block is inspected manually. Being failsafe by itself is not sufficient. Clearly, missed wheels must be mitigated so the system demonstrates high reliability and freedom from nuisance right-side failures In many conventional wheel detectors, a fixed threshold is used to detect a change in the induced signal. If the signal is below this threshold, the event would be missed.
Noise which occurs as spikes can easily cross the threshold and be mistaken as the edge of a wheel waveform.
Employing a variable threshold detection algorithm where the threshold which is automatically adjusted according to the actual signal level generated at that sensor is helpful but may run the threshold closer to the noise floor. Wave shape analysis using the area under the curve is more useful in detecting a wheel. That is, short lived noise will be process out so only waveforms with sufficient area under the curve are considered.

Furthermore, most dual head wheel detector systems use two separate coils, each operating independently with their own magnetic field generation. Each sensor employs a different frequency of operation which leads to cross-talk and inter-modulation as the flux of one coil interacts with the adjacent coil. This creates a beat frequency in the detected signals which varies in frequency and amplitude and is a source of unwanted noise and is difficult to filter if it is within the bandwidth of the baseband signal. To minimize this, the coils are usually spaced far apart which makes the sensor larger.

Some sensor designs multiplex the sensors so only one is on at a time. This removes the crosstalk problem but introduces a delay since each sensor takes time to build up its tank circuit oscillations and limit the rate at which the coils are switched. For fast trains, the sensors do not multiplex fast enough and only a few samples are taken for each wheel waveform. This precludes any kind of digital signal processing because of the insufficient digital sample size.

Dual sensors measure the train speed by measuring the time it takes for the wheel to traverse the sensor, which is only about 9 inches wide. This limits the resolution of the measurement when coupled with a finite sample rate and system noise. Accuracies of 10% are typical at speeds of 90 MPH with typical wheel sensors systems.

Putting two non-vital wheel sensors in tandem creates redundancy, but does not by itself assure a vital system. The use of a plurality of said wheel sensors used in the present invention allows information to be integrated into the control algorithm and allows additional processing and logic to be used on the collective data to determine the actual wheel status.

Most sensors cannot self diagnose a problem, and only detect a problem when a train traverse the detector.

In addition, an accelerometer integrated into the design allows wheel-induced shock and vibration to be measured and detected, yielding additional information such as flat spots, wheel damage or wheels out-of-round. Such wheel defects are reported in correspondence with the wheel count so accurate position of the defect wheel is known.

The essence of this invention is to rectify those problems and provide a sensor and a detection system which includes self test, self-evaluation and mitigation of detection faults as required for vital signaling applications. Herein, we disclose a new sensor design, a new vital evaluation controller and a new track mounted sensor configuration.

Prior Art The authors previous patent, Southon, patent #6371417 describes a wheel counter system which achieves vital safety using eddy current sensors. The coils are frequency multiplexed so the driving electronics is switched between the coils and requires a settling time before the signal is stable. This limits the speed measurement accuracy due to a slower sample rate. For fast trains, only one or two samples for a wheel signal are taken, so any kind of meaningful digital processing of the waveshape is precluded. A fixed amplitude threshold is used which is not adaptive to individual differences in sensor parameters. The Southon patent describes the use of slope detection to determine the slope of the frequency-amplitude curve in order to detect changes in coil parameters. Although providing vital operation, is has limited error mitigation techniques which could overcome wheel misses and avoid nuisance right-side failures.

Shams, patent #633456, cites a simplified oscillator supplied adjustable power from a DAC to calibrate the sensor drive amplitude. It monitors the drive coil amplitudes and compares against a value derived at power up. Fixed amplitude detection is used to detect an off-rail condition which is dependant upon proper calibration. The system does not have an independent method of inducing a magnetic load to test the coils operation. There is no methods to detect if the coil is operating out of tolerance.

Most dual wheel detector in the prior art use two oscillators driving separate coils at different frequencies creating crosstalk between them by heterodyning, also known as a beat frequency.
This beat frequency is variable and unpredictable and low frequency beat components are impossible to filter out because they can occur in the bandwidth of the analog processing circuits.
Generally, the sensors require an adequate separation of the coils, making the sensor larger. The oscillators can be off frequency and is not directly detected, nor is the Q of the coils tested.
Gilcher, patent #5333820, uses a differential bridge to detect wheels but the unit must be mounted on both sides of the rail and needs a potentiometer adjustment to balance the coils. The adjustment is not automatic and subject to errors with temperature, vibration and aging of the components. There is no self test mechanism or discussion of vitality.

Gallender, Patent #5395078 uses a self calibration scheme to continually calibrate the sensor, but cannot calibrate between train wheels hence cannot calibrate out sensor movement slip on the rail. Here is also no self test or diagnostics and the unit operates at a fixed detection threshold. No off-rail detection mechanism is disclosed.

None of the prior art indicates the use of spatial diversity, majority logic and DSP techniques to mitigate the error sources mentioned below. None of the above cited patents acknowledge the sources of error or the mitigating techniques used to correct the errors.

Speed accuracy is not mentioned in any of the cited patents.
Summary of this Invention Several years of extensive field testing with a variety of sensor methodologies has shown the current state of the art in wheel counting precision to be at best, 1 missed wheel per million wheels. By using n independent detectors, the probability of all n sensors missing a wheel is:
Probability of missed wheel = (1 * I Oe-6) n -- equation 1.1 For example, we use two independent detection points to detect the axle so the probability of missing the wheel is negligible (1 in 1012 wheels). To achieve the condition of mutually exclusivity, errors which are common to each independent detection system must be eliminated so what remains are only randomly occurring events which are mutually independent.

To achieve this independence, we employ the concept of diversity, that is different from redundancy. In redundancy, multiple identical processes are used to detect the same event. In diversity, different methods are used to detect the same event. Each of the diverse processes is subject to its own potential error sources, but these are usually different than the errors which may occur in the other diverse process. The probability that both processes will have the same error is greatly reduced since they are programmed differently, run on different hardware and separate sensors. For each axle, 4 waveforms are produced which are processed by a waveshape-based detection means and a majority logic algorithm is used to mitigate wheel waveform anomalies.

To achieve this high wheel count reliability, we use spatial diversity at a single site by placing two detectors at different locations with adequate separation to ensure that a disruptive event at one end of the axle cannot affect the other detector at the other end of the axle at a later time.( for example, a flat spot on one wheel generates a high impact, causing a disturbance in the sensor would likely not occur on the other wheel of the same axle on the other rail sensor some distance and time later. ) Secondly, time diversity is used by making the detections at different times such that a single disturbance at a moment in time cannot affect both measurements.

Thirdly, wheel diversity is used. We measure an axle using both wheels of the axle so that wheel related positioning errors are minimized.

Processing diversity is achieved since each sensor has its own calibration process and detection algorithm, so calibration related and auto-zeroing errors are not common to each sensor.

The sensor The invention employ one drive coil and six detection coils. The detection coils are passive and symmetrically balanced. With an implementation consisting of only one drive coil operating at one system frequency, there is no inter-modulation or beat frequency to contend with. The drive frequency is not multiplexed and the receiver coils are an at all times, resulting in no need for settling time considerations. The sensors are always ready to receive a wheel without delay. The wheel voltage will be generated in real-time regardless of the ADC (analog to digital conversion) sample rate.

Each of the two diverse sensors has independent circuits and process the signals from a different wheel at a different time so that errors introduced at one sensor is mutually exclusive of errors caused at the other sensor.

Since the wheel detector receiver coils are constantly driven, the Analog to Digital conversion rate can occur at a very high rate (ie: 50kHz) so an adequate number of samples from each sensor allows digital processing to be done effectively. The DSP functions and high sample rates allow FF transforms and digital filtering to remove unwanted DC offsets , compensate for sensor slip on the trails and allow calibration between wheels.

The main drive oscillator is driven by either a direct calibrated frequency source or a variable frequency source which varies as the inductance of the primary coil is changed due to the loss of inductance when a wheel is present. Magnetic flux from the drive coil will couple with the coils of the receivers and induce a voltage in them.

The two primary wheel detector coils are positioned between the drive coil and the impending wheel. The wheel reduces the flux available to the receiver coil and causes a drop in the voltage generated. Each coil is independent and creates its own wheel voltage profile.
Mutual coupling between the coils is beneficial and aids in detection when the occupied coil reduces its voltage, it reduces its magnetic coupling to the non-occupied coil that experiences a slight increase in voltage.

Each wheel detector coil is separately detected by its independent a/d channel and analog signal processing.

Each sensor coil voltage is digitized as a single ended input to the ADC. In addition, a differential amplifier is also used to measure the difference in the two wheel detector coil voltages. Separate detection circuits and processor code is used to analyze the wheel event as both two single-ended and as one differential measurement to assure the diversity rules are followed.

The sensor includes a third coil located at the rear of the sensor and as such is heavily damped by the rail. When the sensor moves away from the rail on either side, a rapidly increasing voltage occurs as the dampening influence of the rail is removed. This off-rail detection is reported and forces the system into a failsafe condition. A single coil or a dual coil arrangement may be used.
Two diverse coils are better at detecting an angular movement of the sensor.

The fourth and fifth coils are self test coils used to load the wheel detector coils and simulate a real train wheel. These test coils are looped near the main wheel coils so as to encompass the flux which is passing through the wheel detector coils. The test coils are normally open circuit and are then shorted through a switching means of low resistance to become a load on the wheel coil and create a significant drop in wheel coil voltage. The magnitude of the drop is comparable to that seen by a real wheel presence. Both coils are independently configured for test loads. The coils are activated in a electronic sequence identical to a wheel set of a train.
The timing and amplitude are controllable to high precision and can emulate a wheel waveform at speeds from 0 mph to over 250 mph. The detector coils respond exactly as expected if a real wheel was in proximity of the coils.

All sensor coils are monitored for bandwidth, amplitude and resonant frequency response. This is done in a continual cycle where the main oscillator frequency is used to sweep the receiver coils and determine their frequency-voltage profile. Any change in the coils operation is detected and may result in an immediate fail-safe response.

Recalibration of the sensor wheel detector coils are done on the fly and essentially is the balancing of the two voltages to be similar in value so the difference is zero. When this is achieved, the drive frequency is determines to a value optimum for both coils.
If a balance cannot be achieved, the sensor is deemed unsuitable for use.
The Sensor Placement Geometry The sensor placement geometry is a virtual quadrangle where a sensor is at one corner on a rail while the second sensor is at a diagonal corner on the opposite rail placed longitudinally about 6 feet apart. The strategy is to detect an axle by detecting both wheels with time and space diversity before deciding if a wheel is valid. This configuration does not use the typical wheels-in and wheel-out strategy as seen in virtual block systems because it is not intended to be a virtual block. It is a detection zone of extended length and operates in the systems as a single detection point with vastly improved reliability.

If a single wheel detector misses one wheel in a million and we use 2 diverse sensors, the probability of missing both wheels in two independent sensors, each with its own processor, calibration and spatially diverse position is 1 in 1012.

In addition to the benefits of getting multiple readings from each wheel, the geometry affords the discrimination of different wheel carriage arrangements being able to differentiate locomotive types as well as high rail vehicles, single axle and dual axle bogies.

The speed resolution is improved dramatically as we can now measure the time it takes the wheel to travel 6 feet, instead of 9 inches.

Wheel detector error sources Although vital systems can be built with a single sensor at boundary point of a block, experience has indicated that it is difficult to mitigate all sources of wheel errors with just a single detection point. With two sensors strategically spaced, greatly enhanced wheel mitigation is achieved.

A variety of error source have been known to contribute to the missed wheel event. Some of these are as follows:

1 Rapid temperature changes change the DC baseline value suddenly just before a train approaches. The threshold levels are biased by an additional offset which can reduce or increase or decrease the effective sensitivity of the sensor. This may reduce the noise margin or move the signal baseline further from the detection threshold.

2 Wheel sensors slide on the rail as the rail deflects. This causes a baseline shift which again moves the DC baseline further from the threshold and hence reduces the sensitivity of the sensor. A wheel with a smaller than specified flange depth can be missed. We have recorded the shifting of the sensor under a train which is a gradual shift progressing as the train passes until the final few wheels are missed due to the accumulated baseline shift.

3 Wheel hunting, where the wheel drives around the sensor and is out of range of the sensor. The signal is present but too low to be detected with the current fixed voltage threshold.

4 Direction reversal due to too slow a sampling or multiplexing rate on the coils. It is possible in slowly sampled systems, to detect a very fast oncoming wheel on the second sensor before the first and produce a wheel detection which appears to be in the wrong polarity. Having the second sensor being more sensitive than the first increases the likelihood of this error.

Wheel sensor becoming loose on the rail and vibrating excessively when a train passes. This creates signals which may cause multiple false detections and errors in wheelcount accuracy.

6 Wheels parked on the sensor for long period, even weeks, will cause the calibration circuitry to `zero off the wheel when the wheel is just slightly inside the proximity range of the sensor. When the wheel eventually moves forward, it has reduced sensitivity and may be missed.

7 Dead slow wheels create a slowly changing voltage which must be DC coupled to the detector signal processing and detection logic. Being DC coupled, other low rate-of-change factors such as temperature also cause variations in the voltage which are very difficult to discern from a exceedingly slow creeping wheel.

8 A wheel parked on the sensor for extended periods of time make it difficult to correct for temperature drifting that may occur during that time (and maintain constant sensitivity) 9 Wheel sensors which have no method to self-test themselves for sensitivity and high speed frequency response usually find this failure mode only when a train arrives.
Small flanges on high-rail vehicles fail to trigger wheel detector, or trigger only some units, so confuses the system 11 If the sensor mounted too high on web so that the head of rail dampens the coil enough that low sensitivity results in lost wheels.

12 Sensor mounted too low on the web of the rail so the wheel flange is too far from the sensor. Low signal amplitude results in a higher number of missed wheels.

13 Noise spikes which cross the thresholds and create a false detection.
14 Sabotage or attempts to arm the crossing by placing metal on sensors.

Wheel sensor activation by dragging equipment, such as chains or sheet metal.

height exception report if a flange is detected to be irregularly higher than the norm, or if said flange exceeds a absolute detected voltage greater than the programmed flange height threshold setting for that sensor.

Description of the Drawings The simplified block diagram is shown in figure 1. The main drive coil [8] is driven by a tunable VCO [7]. The field from the main coil is situated so that the produced magnetic field encompasses all other coils in the system, as well as protrudes into the space occupied by any passing wheel flange. As an object enters said magnetic field, it changes the inductance of the main drive coil and changes the voltage generated across the main coil. A peak detector [6]
detects the envelope of the voltage oscillation at the main coil [8] and sends this signal to the Analog multiplexer [4] for eventual conversion to digital codes at the ADC at [3].

The main coil[8] is also monitored by the frequency measurement timer [27]
which reports the changes in baseline frequency to the DSP [1]. Incident wheel therefore produce a voltage and a frequency change, both of which are monitored by the DSP and processed t by the stored control program [28] in the DSP. This creates two diverse methods to measure the effect of a wheel.

In practice, a high speed train will pass very quickly and not allow much time to sample and measure an accurate frequency reading, however, the voltage fluctuations are ample for reliable detection. The frequency detection method is used primarily for low speed detection or wheels stopped or very slow across the sensor. For wheel which park on the sensor for an extended period, the frequency variation is important because it is a physical and absolute measurement, whereas the voltage measurement is relative to the calibration point which may change as the sensor recalibrates from time to time.

The Reference Generator [5] develops a small reference voltage which is proportional to the amplitude of the main drive coil[8]. This is used as the reference voltage for the ADC[3] so that all measurement are ratiometric with the excitation voltage. This tracks amplitude changes in the drive coil and cancels out the affect so as to keep the digitized values constant. In essence, voltage fluctuations in the drive circuit are nullified.

The main drive VCO[7] is tuned by a series of capacitors, eg. [26] which are switched in and out of the VCO circuit by mosfet switches [25] in order to alter the frequency of the VCO. This is done in such a manner that the amplitude and pure sinewave excitation is maintained over all the tuning range of the VCO[7]. As many capacitors[26] and mosfet transistors as required for the tuning precision can be used so that very precise control of the frequency of the VCO[7] is achieved. The preferred embodiment, 8 stages are used to create 256 discrete frequencies. The reader will recognize that more or less discrete frequencies can be easily obtained.

The process can read the frequency and voltage of the main drive coil as stated. The tunable VCO[7] allows the processor to scan the frequency range of the main coil to determine the amplitude vs frequency response of the coil. If the -3db frequency points are determined, the bandwidth of the coil can be computed. This bandwidth is a function of the Q
of the coil and allows the DSP to determine the quality of the main drive coil to determine if it is within the safe functional range. During this test, the ADC[3] is switched to an internal reference source so that absolute voltage measurements are used during the frequency sweep.

The VCO[7] is also tuned periodically to account for temperature and aging of the system components.

The primary reason to examine the drive coil and its circuitry in such an extensive manner is to determine its health and ability to safely perform. The main drive coil is not used to detect wheels, but is used in the event of a dead slow or parked wheel on the sensor.
In this condition, the amplitude , frequency and Q of the coil all indicate the presence of a metallic object near the sensor drive coil.

In summary, drive coil voltage, frequency and bandwidth are all measured to determine if the coil is fit for vital duty. Any defect in any component of the VCO and its calibration components will yield unacceptable variations in the drive coil performance and will shutdown the sensor.

The wheel receiver coils at [10] and [12] are aligned to half-lap the main drive coil and to be situated between the coils and the wheel detection area. These receiver coils are tuned to operate at a frequency above the main drive frequency and also below the main drive frequency. The coils are equipped with envelope detectors which present the voltage to the differential amplifier [9] which removes the common mode voltage and amplifies the difference voltage between the two receiver coils. Since the receiver coils [10],[12] are symmetrically mounted over the main drive coil[8] and are symmetrically tuned with respect to the frequency of the main coil, the difference voltage is theoretically null because the system is balanced magnetically and electronically. It is the function of the calibration algorithms in the control program [28] to adjust the VCO [7] slightly in either direction so that the difference signal is null. Failure to balance the symmetry of the coil arrangement is an indication of a defective magnetic balance and the compromised system is detected.

This magnetic balance is shown in the frequency amplitude plot of figure 2.
The receiver coil A
[1] is below the main drive coil frequency [2]. Receiver coil B has its frequency [3] operating above the main drive frequency [2]. The main drive coil is tuned to be between these two receiver coils in the approximate centre, although system tolerances usually require an offset in the frequency in order to achieve this balance.

As a wheel approaches receiver coil A [10] back in figure 1, the resonant frequency of the receiver coil is raised due to the eddy current affects discussed earlier.
This cause the resonant frequency of the driver coil A [10] to rise, moving farther from the frequency of receiver coil A
and becoming closer to the resonant frequency of receiver coil B. Hence the voltage [29] will fall and the voltage [30] will rise. The onset of the wheel over receiver coil A
unbalances the magnetic symmetry of the sensor and create a net negative voltage output from the differential amplifier. The onset of a wheel over sensor B will cause the reverse unbalancing and create a positive differential voltage.

Note also that the absolute value of each receiver coil output is also digitized by the ADC, so that the differential calculation can be done is software as well. The differential amplifier is a diverse method of detecting the degree of imbalance and allows for a confirmation of all A/D, multiplexing and processing accuracies. In addition, the differential amplifier can be used to introduce gain into the system to increase overall sensitivity if needed.

The receiver coils are tuned tank circuits which respond to the frequency and amplitude of the flux of the magnetic field surrounding the main drive coil. The influence of the wheel causes an unbalance in the fields symmetry which the adjacent coil can detect. When one coil detects a magnetic disturbance, the other coil responds in the opposite polarity because the common drive flux which permeates both coils is affected.

It is within the scope of this inventions apparatus and software to drive the main oscillator with a constant frequency source which is stable during a wheel event and to measure the changes in the induced voltages at each receiver.

Referring to Figure 1, the Test coils A [18] and test coil B [19] are adjacent to their respective receiver coils. When these coils are shunted to ground, they create a magnetic loss which can be adjusted by the load resistor to generate an effect equivalent to a very small wheel. This shunting is controlled by the DSP [1] and is used to emulate a train wheel at various speeds. The DSP will create a steady state load to measure the sensitivity of the coil to determine if the sensor is operating in range. The DSP [1] can also create a sequence of test coil actuation that emulate a wheel passing, or emulate a series of hundred of wheels at high speeds. In this manner, the high speed response of the sensor can be verified without needing a train present to validate the system. These tests can be run automatically in a non intrusive distributed fashion or when commanded by the system evaluation module which has visibility within the system to know precisely when a train is no longer near the sensor being tested.

The off-rail detector coils [16], [17], are mounted at the rear of the sensor, right tightly up against the web of the rail. These coils are tuned broadly over the range of the main drive coil [8] to produce a small voltage since the coil is heavily damped by the rail. If the sensor becomes loose and moves away from the rail, the off-rail detector circuits produce a sharply increasing voltage which is detected and used to shutdown the system before any fault can occur.
As part of the regular vital system health monitoring, these off-rail voltage are constantly being tested.

A scan of the VCO [7] and drive coil frequency through the range of each receiver coil also measures the voltage, bandwidth and Q of each receiver circuit and is used in a similar fashion as discussed for the main coil and is an ideal way to characterize and measure the health of the receiver coils.

The off-rail detector coils are also swept in frequency to measure the voltage and Q in a similar fashion to verify performance. Of rail detection coils are heavily damped and have a low Quality factor. Once the sensor is moved away from the rail, the voltage increased rapidly, indicting a removal operation. Removal of the sensor by I/4 inch or less will trigger the off-rail detection circuit.

Incident wheels create a voltage change at each respective receiver coil as discussed. Figure 3 shown the waveform produced when a wheel passes at any speed over the sensor.
The voltage first goes positive while entering the sensor, then returns to zero when the wheel is centered, then dips negative when leaving the sensor. The polarity of this waveform determines the direction.

It is important to note that for a steady occupancy for an extended period of time, that the system has the main coil frequency and amplitude and Q, as well as the receiver voltages and Q
measurement to indicate the presence of a parked wheel over the sensor. This yields positive and accurate detection without drift and recalibration problems associated with earlier sensor technologies which use simply the relative voltage across the coil.

Note that in the preferred embodiment, that planar magnetic coils are used which are built with PCB traces and are air-cored. An implementation of the symmetrical magnetic sensor achieved with ferrite or other high permeability magnetic material is within the scope of this patent.

The air core has a more uniform flux density around the coil because the permeability is constant.
Ferrite based coils have high flux densities in the core and reduced flux at away from the core which is primarily leakage flux.

An internal table in Flash Memory [2] holds system limits and real-time and historical performance results. The Flash provides a lookup values for main drive coil frequency versus measures sensor temperature [31 ]. This is used to compensate for the temperature induced frequency changes in the oscillator and renders the compensated frequency measurement temperature independent and thusly, dependant only on the occupying wheel.
Over the life of sensor, temperature readings versus oscillator frequency are constantly being compared to the table entries allowing the opportunity to update the tables at specified times. This can be done during the yearly vital test and re-qualification of the system.

Figure 4 illustrates the concept of the vital Primary Detection Quadrangle (PDQ) strategy.
Although vital systems can be built with a single sensor at each end of a block, experience has indicated that it is difficult to mitigate all sources of wheel errors with just a single detection point. With two sensors closely spaced as shown, greatly enhanced wheel mitigation is performed. In addition, the extra visibility to the axles of a train greatly increased the data from which to draw logical conclusions and to recover safely from missed wheel events.

The Primary Detection Quadrangle is a virtual quadrangle where a sensor [1] is at one corner while sensor [2] is at a diagonal corner. The strategy is to detect an axle by detecting both wheels with time and space diversity before deciding if a wheel is valid. Using basic rule based logic helps decide if a wheel is valid, while majority logic is used as a tie-breaker.

The spacing of the sensors is important if bogie classification and error correction is desired. The evaluation of the data from both wheelcounter is done by the Vital Zone Evaluator [5] . Each axle is sensed 4 times, utilizing both wheels so that errors introduced by one wheel are mutually exclusive of errors caused by the other wheel. The VZE can handle 4 sensors with the distance to each sensor limited to a cable distance of 5000 feet. The VZE also supports a dual radio system [14] and [15] which allow data to be sent upstream and downstream limited by only the range of the radio system.

The VZE also support two vital Inputs and Two Vital Outputs for collecting and distributing information to other points in the network.

Detailed Discussion of the Invention Eddy current induced losses when a wheel flange is in the proximity of the generated magnetic field will lower the effective inductance of the oscillator coil and increase the frequency. The baseline voltage can be stored in the processor for reference. As the auto calibration zeros off any small residual voltage offset, it is possible to zero off a wheel a bit at a time until eventually the sensor is at a new calibration point. It the changing voltage baseline is due to temperature, then we have no problem because the new baseline be valid and the relative threshold will be correct.
If the changing baseline is due to a creeping wheel, then the zero-off is detrimental, because it is removing some of the wheels effect and hence reducing the sensitivity of the sensor to the remaining wheel portion.

The frequency shift is also another method of detection of the baseline. The frequency will be shifted when an object is placed in the magnetic field. The shift in frequency also has a temperature dependency, which can be partially compensated for by measuring the coil temperature and correcting for the temperature shift.. This data is also gives additional insight into the wheel presence but is not deterministic by itself.

The baseline shift discussed is a factor of the sensor being mounted across several inches of track and being affixed by bolts. Friction holds the sensors in place. A passing train can cause considerable deflection of the track and cause the sensor to slide and wear at the contact points on the rail. Some sensors use a small metal base to affix the sensor to hold it out from the rail so that the mount constitutes a single point mount. However, the sensor is more prone to vibration if it protrudes too far from the rail. Also, off rail detection is not possible if the sensor is physically not in direct contact with the rail An extended occupancy detection occurs when a sensor has been occupied for longer than 20 minutes, This sets an internal flag which will cause the sensor to recalibrate immediately when the wheel leaves the zone. The sensor will then zero off a wheel and maintain the new calibration point. When the wheel finally moves after an extended occupancy, the voltage will move in the opposite direction to that seen when a wheel enters, and an immediate recalibration is triggered, recovering quickly from the extended occupancy. Since only one sensor of the pair is occupied by a wheel in this fashion, the other sensor acts as a reference to indicate true wheel count so the wheelcount remains true even after such an extended occupancy.

As the wheel rolls across the sensor, the mutual coupling between the primary driver coil and the wheel coils varies and the driver coil pick up the coupling from each primary coil in turn. The output from the differential amplifier is a sine wave which swings positive, back through zero and negative and back to the unoccupied steady state value. The wheel is known to be centered over the sensor body when the output goes through the zero crossing. A centre of zero flag is set internally which can be polled by the evaluator if needed during low speed testing. The centre of sensor LED will also light if enabled to aid in system testing and validation.

Since the measurement is differential, any changes in the amplitude of the excitation that occurs in both primary excitation coils cancels (nullify) each other. This may be true only if we use ratiometric operation, using the voltage from the drive coil as the reference voltage.

The DSP can process these at a continuous rate to remove induced noise spikes, extract the baseline shift dynamically, recalibrate between train wheels to eliminate accumulative effects and use waveshape detection algorithm instead of a fixed or variable voltage threshold for the detection.

The PDQ approach also provides for more accurate time and speed measurement, because we take the time to traverse 6.0 feet, rather than 5 inches. The resolution of the speed measurement is improved by having a longer period in which tosample the wheel and by the higher sampling rate of the system, such that the speed can be measured to better than 0.1 % at 200 MPH.

For vital equipment, self-testing is necessary to assure that the system is capable and safe.
We build in a self test magnetic circuit which loads the primary coils with a known amount of load, equivalent to a wheel of the smallest influence. This load is totally magnetically independent on the primary coil and is electronically activated at a DC level to measure the sensor calibrated response, and again run at high speed to generate a pattern of magnetic disturbance to emulate a high speed train. In this way, the operation of the sensor can be measured and verified again safe limits and kept in service knowing it performs perfectly.
This self test can be run at any interval and is usually interleaved with regular detection algorithms which run at a reduced speed momentarily so as to be still sufficient to detect an oncoming train at full rated speed. Upon detecting a train, the self test is immediately terminated and scanning of the coils continues at the normal rate. That is, the sensor is never allowed to remain offline for more than 32 usec at a time.

Processing Algorithms The sensor configuration on the track as shown in figure 4. This arrangement mitigate all sources of errors cited above, non of which are addressed in prior art current technology.

Using a high speed DSP processor, A/D conversion of the sensor signal can take place at a rate of over 200,000 sample per second. The DSP system will process the waveshape of the wheel induced waveform by using digital low pass FIR filtering, impulse noise rejection and signal integration to determine a successful detection.

Another processing method performed by a statistical correlation between the incident wheel waveform and a known good detection waveform from memory (but speed corrected). Suspect wheel waveforms are keep memory for analysis by the correlator logic and fuzzy logic analysis such that suspect waveforms can be adjudicated and used in the voting process even though they are not ideal. For example if two good waveforms and two suspicious waveforms where acquired, the suspect waveforms would be subject to a further analysis by waveshape and by the duration of the pulses and period between the pulses that would show that something was detected and is the proper duration with respect to the good wheel detection and of significant amplitude to be considered a valid wheel when taken in context of the other 3 measurements Questionable detections are hence further processed to determine the actual wheelcount.
Generally, significant wheelcount waveform deviations only occur only is isolation and three remaining wheel waveforms would provide the correct logical decision.
Questionable waveforms are counted and used to indicate the overall status and health of the sensor.
Wheel coils which are generating larger numbers of questionable detections can be identified and replaced before any loss of service or reliability occurs.

The sensors are magnetically balanced and electronically balanced. Once in balance, they produce the same deflection amplitude. A properly balanced and calibrated coil that passes the self test has very low probability of missing a wheel. Wheel sensors which report consistently lower amplitudes are threshold - adjusted and compensated by the control software to account for the lowered sensitivity. The said self test coils can provide health monitoring and provide a fail-safe restrictive state to be applied well before an actual train appears.

Rule-based Logic for Error Mitigation There are many logical deductions which can be applied to mitigate a large number of missed wheel events. The deduction logic is based in part in many axioms, some of which are stated below as example:

If a wheel enters a single sensor ( with dual coils) passed at some significant value of speed, and we measure the first wheel but miss the second due to some unusual disturbance, it is safe to say that the train did not stop immediately and the wheel detection may be deemed valid even if the second detector showed its respective waveform to be abnormal, but was in the correct phase and timing, we can conclude the wheel was properly detected.

Conversely, if a wheel fails on the leading sensor, but passes on the trailing sensor and is traveling at a significant rate of speed, we can say the wheel detection is successful.

Also, we know that a train cannot stop and reverse instantaneously. If a reverse direction wheel is sensed on a forward moving train at a significant threshold speed, we can conclude that the train is going in the same direction and reverse the sign of the wheel count to correct the error.
However, This error occurred on prior art sensors that used 6 inch spacing between detection coils. The current invention uses a 6 ft nominal separation between the sensors so speed and direction are resolved unequivocally.

Majority logic can conclude that if 3 out of 4 coils report a good wheel and the remaining coil is suspect, then we can conclude a good detection.

If a dual axle bogie is seen successfully leaving the PDQ but was not accurately measured coming in, we can safely assume the PDQ is cleared and count the exit wheel events only.

These know facts among others can be used to mitigate wheel errors if they occur. The deduction process is achieved primarily on the fact that we get more information from the diverse pair of sensors and can draw these conclusions safely.

For extended occupancy, we detect the wheel by several factors:

1. One or both coils will have a voltage lower than the unoccupied value. The magnitude of the change is dependant upon how close the wheel is.
2. One or more detector coils will be occupied for an extended period, defines here as 10 minutes or more.
3. The Q of the detector coils will be lowered.
4. The bandwidth of the driver circuit is higher, and the Q is lowered.

5. The resonant frequencies of the receiver coils will be raised.

6. The calibrated response from the test coil will be less showing the coil(s) are occupied.

7. The sensor measured parameters are compared against the other sensor in the diverse pair, and such factors as temperature drift which is common for the apir, can be subtracted off.

If the wheel is marginally proximal, the detection may be minor and below the threshold, so the unit will zero-off the wheel. This may occur several times until the maximum allowed offset correction is reached and the sensors goes occupied.

When a zeroes off wheel eventually leaves the sensor, a large negative waveform uniquely exists, which is immediately detected and a new incremental calibration done immediately, bring the sensor into proper calibration for all subsequent wheels.

Sensor construction The completed sensor has 6 coils situated strategically to perform various duties. In the preferred embodiment, planar coils made on a PCB are used to provide very accurate reproducibility of the coils and maintain very stable distances between them. The use of ferrite coils is also another embodiment of this invention as is recognized by anyone familiar with wheel sensor inductive technology.

The use of air cores reduces the weight and raises the mechanical resonant frequency of the sensor so as to be far above frequencies encountered on any vibrating rail.
This fact prevents standing waves from being generated when the unit is subjected to vibration energy at its resonant frequency, that is potentially destructive to the internal components. Air cores are also more predictable in their characteristics not being reliant on the permeability of the ferrite core.
Also, ferrite is very brittle and may be broken with high shock values encountered with track related forces and shock. Very heavy ferrite will tend to put additional mechanical stress on the mounting hardware and resins and may tend to loosen its supportive structures somewhat over years of vibration. This will introduce noise as distances between magnetic components can vary with vibration. Also, stray fields and coupling between adjacent coils is a concern, as flux lines will tend to follow the high permeability path through the ferrite cores rather than flow out to the detection zone.

The air cores in contrast are very thin, and the entire sensor can be made smaller, lighter and able to be mounted on lighter gauge rail. Ferrite designs need to be much larger and wound coils have looser tolerances than printed circuit coils.

Also, ferrites have a temperature co-efficient which changes inductance of the coils with temperature, so that at low temperatures, the reduced inductance changes the performance of the system. With air cores, the permeability of the magnetic system remains fixed and predictable.
State Machine driven software Using a finite state machine, the states are driven by the detection of the first wheel on the first coil of the sensor, followed by a zero crossing, followed by a detection on the second coil followed by a return to the normal unoccupied state. The `Center of sensor' LED is lit to help in manual testing of the wheel detector unit. LEDS also exist for `coil A
Occupied', and `coil B
Occupied'. These LEDs are useful during testing and commissioning but will timeout after 1 hour of power application so as not to attract unwanted attention. The signalman can access the system and turn on all LEDS at any time to aid in diagnostics.

A valid wheel event must drive the state machine through all four states to be considered a valid wheel. The state machine accounts for the reversal of a wheel anywhere over the sensor and counts the number of axles seen.

The state machine variables also starts and stops the gated timing reference frequency and produces the transition time between the separated sensor heads. Although the sensors can be used as standalone devices, the integration of data from the aggregate pair is advantageous as described for error mitigation.

With the speed of the train known, the time between wheels can be easily measured and calculated so that a minor calibration can be done between axles or bogies.
This only takes a millisecond to read the unoccupied sensor values and do a re-balancing of the magnetic circuit. In this way, the offset which occurs gradually or abruptly while a train passes can be detected and adjusted out so as not to be accumulative between axles. This compensates for sensor slip on the rail as the rail flexes and sensors move on their mounts.

The communications controller has a port available for a laptop computer to be connected so that all system data for all of the wheel detectors can be displayed and examined.
This helps diagnose the entire system from the warmth of the bungalow. With an internet connection, the signalman can monitor the wheel system from the service center, control room or almost anywhere. High speed data logging can be enabled to allow discrete digital sample data to be output to an external device for analysis and graphical analysis off-line. A compression algorithm is used to compress out most of the unoccupied sensor readings which occur even when a train is passing over the system.

Self Test Processes The following parameters are tested in real time by the control firmware:
1. Coil absolute voltage limits 2. Receiver wheel coil balance not at null (zero).
3. Drive coil and receiver coil bandwidth.
4. Amplitude variation with application of test coil.
5. High frequency response of sensor with high rate of test sequences 6. Frequency of main driver coil.
7. Coils tested for voltage - frequency response and Q factor.

8. CPU clock speed ratio test (ratio of diverse clock sources) 9. CPU software tests for internal execution errors.

10. CPU tests for internal memory errors, failed CRC checks and watchdog timer resets 11. Off rail detection coils tested for functionality and Q and amplitude.

12. Noise threshold tested by application of test coils for bursts-should not be detected 13. Extended train tests where several thousand wheels can be emulated and verified.

14. Low power supply voltage 15. Sabotage protection - application of improperly timed activation sequences are ignored 16. Excessive non symmetry of wheel waveforms 17. Excessive communication CRC errors and communication dropouts.

18. Sabotage detection by excessive signal voltage on each coil consistent with an object placed on sensor face ( fouled sensor detection) 19. The execution loop time of the program and testing of the control flow 20. Internal RAM and program Flash memory self tests 21. Testing the microprocessor configuration registers for inadvertent changes.

22. Vital variable and vital timers in the control code have diverse internal representations.

23. Vital variables have limited life span and expire quickly, so stale data is eliminated.

It is important to note that all of the above tests are done without the need for a train.

These tests are done in a distributed manner as not to affect the sensor system capability to detect a train. Once a train is detected, all self testing is suspended until the train passes.

All of the above conditions generate error codes which can be reviewed at a remote station for diagnostic and maintenance scheduling.

Claims (34)

1. An enhanced wheel sensor unit,(also known as a wheel detector, wheel presence detector, axle detector or axle counter ), that senses wheel speed, direction, wheel count, sensor rail position and sensor temperature, comprising a single driver coil; a pair of coils used to detect wheels; a pair of self-test coils; and a pair of coils to detect if the sensor is on the rail; a high speed digital signal processor; a communications port; a vital evaluation unit; a power supply; and software methods of calibration, monitoring and self- test.

2. The vital evaluation unit of claim 1, which communicates with a multiplicity of said wheel sensor units and evaluates the combined information from these sensors in a vital manner in order to provide at least the following data: wheel count, train direction, train speed, train length, car counts, car classification, track occupancy, head of train, end of train, and train progression into a block.

3. The vital evaluation unit of claim 1, wherein the timing of the train wheel passages are determined so that individual calibrations of the sensor can be done in real time between the axles of the passing train and thereby eliminating accumulative errors due to sensor movement under a train.

4. The vital evaluation unit of claim 1, wherein the invention uses space and time diversity by mounting a sensor on both rails separated by a distance representative of slightly longer than by a bogie on a typical car. Each sensor itself is comprised of two detection points, yielding 4 independent measurement points for each axle.

5. The vital evaluation unit of claim 1, wherein deductive and majority logic will process diverse measurements of a single axle and produce the valid wheel data even if any one or more readings being corrupted.

6. The vital evaluation unit of claim 1, wherein virtual train emulation is performed by train emulation software that generates train wheel flux variations using the independent test coils by causing magnetic flux variations of similar magnitude as real wheel flux changes, so as to test the entire sensor and subsequent processing and control systems as if a real train had arrived. Trains of arbitrary length and varying speed can be emulated by downloading to the sensor controller, a test file containing train speed, consist makeup, wheel radius, carriage spacing, bogie spacing and inter-car spacing and braking and acceleration profiles of the test train.

7. The virtual train emulation of claim 6, wherein said test files may be controlled by remote commands received from the wayside processors which direct the test train to change speed, or come to a stop utilizing the stored train braking and acceleration profiles contained in the test emulation files such that real train responses are properly emulated.

8. The wheel sensor unit of claim 1, wherein a networked array of wheel sensor placed in a specific arrangement of at least two diagonally opposed wheel sensor units on opposite rails forming a vital detection quadrangle used to detect one or more train axles on ingress and egress at the boundaries of said vital detection quadrangle.

9. The vital detection quadrangle of claim 8, wherein 3 sensors are used on the 3 access points of a track switch to provide vital train detection regardless of the switch settings such that the train may traverse in either direction over the switch and maintain accurate train data.

10. The vital sensor unit of claim 1, wherein vital interlocking means will preclude the switch from activation while said train is within the vital detection quadrangle, said interlocking being performed by at least one vital hardware relay or at least one solid state control relay wired in to the switch control means.

11. The vital sensor unit of claim 1, further comprising at least one solid state vital input port to read track switch contact points and reporting said indication status in the communication protocol with the wayside communications equipment.

12. The wheel sensor unit of claim 1, wherein the sensor is capable of zero speed and bidirectional operation of wheels and bogies in any sequence over the individual sensors or over the primary detection quadrangle regardless of multiple stops or reversals over any time period.

13. The vital detection quadrangle in claim 8, wherein the having a distance of separation being slightly greater than a dual carriage (bogie) such as to contain both axles between the sensors such that other carriage types with different geometries generate a different sequence which can be differentiated from the standard bogie for the purpose of train/car classification.

14. The vital detection quadrangle in claim 8, wherein said dual sensor system acts like a highly reliable single point wheel detection in the overall larger block control and train detection system.

15. The primary detection quadrangle of claim 8, wherein the speed is measured by accurate timing of the time taken for a wheel to traverse the boundaries of the said primary detection quadrangle.

16. The wheel sensor unit of claim 1, wherein said sensor may communicate as a single standalone unit in a point to point communications scheme,

17. The wheel sensor unit of claim 1, wherein the sensors are networked on a multiple access bus and addressed individually through a software protocol carried by said network.

18. The wheel sensor unit of claim 1, whereas a single tunable drive oscillator is used to excite all receptor coils , thereby eliminating multiple oscillators and associated inter-modulation distortion and heterodyning effects.

19. The wheel sensor unit of claim 1, wherein at least one discrete off-rail detection coil is used to detect the proximity of the rail in order to measure the distance the sensor is from the rail and provide early detection of the sensor removal, sabotage or loosening of the sensor mounting hardware.

20. The wheel sensor unit of claim 1, wherein a pair of self-test coils exist in order to magnetically load teach sensor coil independently and in a specific sequence and frequency to determine the static and dynamic response of the overall sensor sensitivity and response and determine its state of operability and safety at zero to high speed wheel events.

21. The wheel sensor unit of claim 1, wherein the main oscillator sweeps its resonant parallel RLC tank circuit to determine the resonant frequency point and measure the voltage, bandwidth and Q of the primary drive coil for the purpose of determining the operability and safety of the calibration of said main drive coil.

22. The wheel sensor unit of claim 1, whereinthe main oscillator sweeps the receiver coils in frequency to determine the resonant frequency point and measure the voltage, bandwidth and Q response of each of the individual wheel detection coils and off-rail detection coils.

23. The wheel sensor unit of claim 1, where the resonant frequency of the main drive coil being tuned to be at or near the resonant frequency of the receiver coils . In the preferred embodiment , the primary drive frequency is operate just below the common resonant frequency of the receiver coils. In alternate embodiments, the drive frequency may be adjusted to be between the resonant frequencies of the two receiver coils.

24. The wheel sensor unit of claim 1, the continuous generation of drive magnetic field and the continuous stream of sensor a/d conversions at high speed free from settling time, switching and multiplexing concerns.

25. The wheel sensor unit of claim 1, wherein the processes the waveshape of the wheel induced waveform to determine that is has correct attributes in correlation with the other wheel sensor signals to detect the presence of a wheel. The preferred embodiment providing a detection means proportional to the area under the voltage-time curve derived from the sensor, plus a means to filter and eliminate noise spikes from the calculation

26. The wheel sensor unit of claim 1, wherein the signal detection algorithm is adaptive, being adjustable based on test results from the internal self test coils so that the sensor calibrates and adapts to individual sensor parameters to account for manufacturing tolerances, changing field conditions, rail gauge, variable rail magnetic parameters and component aging processes

27. The wheel sensor unit of claim 1 where the detection means is adjustable based on historical data from previous wheel detections so that the calibration point is adaptive to different individual sensor parameters, rail wear, flange height variations, rail size, as well as changing sensor changes due to temperature and aging processes.

28. The wheel sensor unit of claim 1 where the timing of the train is determined so that individual calibrations of the sensor can be done in real time between the axles/bogies of the passing train and thereby eliminating accumulative errors due to sensor movement under a train.

29. The wheel sensor unit of claim 1, wherein the system can be selectively programmed to accept or ignore high-rail vehicles which are detected by the unique sequence of activation of the dual sensors in said detection zone.

30. The wheel sensor unit of claim 1, comprising 6 coils planar magnetic printed circuit board coils or ferrite cored coil construction. The preferred embodiment using printed circuit board planar magnetic coils.

31. The wheel sensor unit of claim 1, providing sabotage resistance by requiring a sequence of activation at two locations longer than an arms length separation and further requiring a consistent and precise timing between all four wheel detections.

32. The wheel sensor unit of claim 1, further comprising an accelerometer means for measurement of single dual or triple (X,Y,Z) axis wheel induced vibration and shock.

33. The accelerometer of claim 28, wherein wheel defects ( flat spots or broken wheel) which cause rail vibration high impact load events are measured and recorded and attributed to the offending wheel specifically, said events being used for real-time alarming and said events included in a wheel impact report used for offline analysis.

34. The wheel sensor unit of claim 1, wherein the close proximity of a particular flange in correlation to the average flange detection voltage, will cause a real-time alarm and flange