Field of the Invention

[0001] The invention relates to monitoring the condition of vibrating structures, or structures that are otherwise subjected to cyclical loads. The invention is particularly suited to monitoring defect conditions in vibrating screens units for mineral processing.

Background of the Invention

[0002] Structures and equipment subjected to vibration are often at risk of fatigue failure. The cyclic loading can propagate cracks through components until the part eventually fails (fractures). Damping the vibration can reduce the risk of failure however in some structures, such as vibrating screens for material processing, vibration is an inherent necessity.

[0003] Vibrating screens are used in a range of industries to sort particulates into different sizes. Coal processing plants often use vibrating screen units to sort coal into size categories and to dewater or drain medium from processed products. These units may have a single screen deck or several screen decks extending between two side plates. The vibrating screens agitate the coal so pieces smaller than the screen apertures pass through while coal remaining on the screen moves downstream, often to a subsequent screen for further separation. In dewatering and medium drain applications the coal remains on the screen surface while the surface water or the medium passes through the screen.

[0004] The screen decks may be 1 metre wide by 2 metres long, up to about 3 metres wide by 6 metres long, although even larger dimensions are possible. Normally, the screen decks are rigidly mounted to the side plates which are supported on spring mountings. Beams attached to the side plates equally spaced along the length of the screen support 'screening media'. Screening media is normally a set of replaceable panels, typically about 1 ft by 2 ft in size. The majority of modern screening media are polyurethane but some are still stainless steel. These panels are slotted and fix into rails mounted on the beams attached to the side plates. Other screen deck designs are possible but the above configuration is predominantly adopted. A vibration generating drive engages the side plates. The upstream end of the unit connects to a feed chute whilst the downstream end feeds a discharge chute.

[0005] The vibration drive system has a variable output to vibrate the screens at a desired frequency and amplitude. Any resonant frequencies of these screen units are avoided to guard against excessive displacement. Similarly, the screen motion during vibration is usually controlled such that it is linear and generally at 45 degrees to the horizontal or vertical (e.g. reciprocating up and down). However, in some cases the screen motion is deliberately elliptical or circular.

[0006] Failures of the screening media support structures (i.e. the rails and/or beams discussed above) or the side plates disrupt coal processing for the duration of the necessary repairs. Unscheduled plant shutdowns such as this have large cost implications related to the loss of production and follow-on effects.

[0007] To prolong the operational life of vibrating screens, the operators will typically endeavour to avoid running the vibration generator at or near the resonant frequency of the screen unit. As discussed above, the deflections and stresses at resonant frequencies are relatively large. Unfortunately, the resonant frequencies of vibrating screen units tend to shift over time, possibly due to crack propagation in some components or loosening in some joints or threaded fixings. This can bring the operating frequency close to a resonant frequency which adds stress to the screens, in turn increasing the rate of crack propagation, thereby further changing the resonant frequency and further increasing stresses. Systems such as that described in WO 2009/02606 (Metso Mineral Industries Inc.) use ' orbit analysis' (the elliptical or circular path of the screens in a single plane) to monitor the displacement at selected points within the vibration screen units during operation. An increase in the orbit is taken as a shift in the resonant frequency of the unit and used for feedback control of the operating frequency to keep away from the natural frequencies of the structure.

[0008] While the system may extend the operating life, it does not avoid the eventual fatigue failures in the screens. The failures may be less frequent, but nonetheless disrupt the operation of the plant during repair or replacement of the screens.

[0009] Any reference herein to a patent document or other matter which is given as prior art is not to be taken as an admission that that document or matter was known or that the information it contains was part of the common general knowledge as at the priority date of any of the claims.

Summary of the Invention

[0010] With these issues in mind, the present invention provides a defect monitoring system for a vibrating screen unit for processing particulate material, the screen unit having at least one screen deck surface defining an array of apertures to sort the particulate material by size as the particulate material moves over the screen deck surface in a feed direction, the defect monitoring system comprising:

at least one motion sensor for sensing motion of the screen in a direction transverse to the feed direction and parallel to the screen deck surface; and, a processor for receiving output from the at least one motion sensor to determine displacements, at predetermined time periods, in a direction transverse to the feed direction and parallel to the screen deck surface, and auto correlating the displacements over a time interval, comparing the autocorrelation with predetermined reference data indicative of at least one defect condition to detect the defect condition and estimate an extent to which the defect condition has progressed toward failure.

[0011] Throughout the description and claims of the specification, the word "comprise" and variations of the word, such as "comprising" and "comprises", is not intended to exclude other additives, components, integers or steps.

[0012] The motion at selected points in the vibrating structure can be characteristic of particular defects. Furthermore, the output from a motion sensor may also be characteristic of the extent of the defect which effectively provides an estimated timeline to failure. By compiling reference data from screen units in a known good state, and at varying (known) degrees of defect progression, a database of 'signatures' for particular defects, at various stages, is generated for comparison with future displacement data. Using this technique, a current defect condition can be monitored in 'near real time' during operation. Trends in performance over time may be identified and, the onset of failure is foreshadowed.

[0013] The Applicant's work in this area has found that monitoring the displacement in one particular direction is more efficient than monitoring and processing displacement in all 3 axial directions. Analysis of the displacement in the direction transverse to the feed and parallel to the screen surface, provides an earlier indication of a defect, and requires less processing power than systems monitoring 3-dimensional motion. In this field, it is conventional to apply a Cartesian frame of reference in which the X-axis extends in the general feed direction, the Y-axis extends generally upright and orthogonal to the X-axis, and the Z-axis is mutually orthogonal to both and extends across the screen deck, transverse to the feed direction. Normally, the sensors used on vibrating screen units are tri-axial accelerometers and these are installed to align with the X, Y and Z axes discussed above. Under normal operating conditions, the motion in the Z-axis is substantially less than the X and Y axis motion. The reciprocal or elliptical orbit of the screen deck and the mass of the particulates feeding into, though and exiting downstream of the screen mean vibrations in the X-Y plane are greater. Z- axis displacement of the screen unit is smaller and less variable in normal operation. If fatigue cracking initiates in the screen unit, the Z displacement may change, but the difference is often too small for reliable detection.

[0014] Autocorrelation compares the output signal from a single sensor with a time delayed copy of itself in order to reduce the level of noise in unprocessed output signals. However, autocorrelation of the Z displacement signal, not for the traditional purpose of signal noise reduction, but for the purpose of change detection, provides a substantially earlier and more reliable indication of a defect condition than monitoring 3-dimensional motion, regardless of whether autocorrelation and/or cross correlation is used.

[0015] Once autocorrelation of the Z-displacement reveals an early indication of a defect, the system can then consider the X and Y displacements and various autocorrelations and cross correlations of these signals to better predict the nature and location of the defect. Monitoring the defect for a longer time (because it was detected earlier) allows more data collection for more precise identification of the defect characteristics using the reference data. Skilled workers in this field will understand that an accurate timeline to failure and the likely location of the defect, significantly reduces maintenance downtime and disruption to the processing operations.

[0016] In another aspect, the present invention provides a method for monitoring defect conditions within a vibrating screen unit for processing particulate material, the screen unit having at least one screen deck surface defining an array of apertures to sort the particulate material by size as the particulate material moves over the screen deck surface in a feed direction, the method comprising the steps of:

attaching at least one motion sensor at a respective position on the screen unit; processing an output from the at least one motion sensor to:

determine displacements, at predetermined time periods, in a direction transverse to the feed direction and parallel to the screen deck surface;

determine an autocorrelation of the displacements determined over an autocorrelation time interval, with the displacements determined from a preceding autocorrelation time interval;

comparing the autocorrelation with predetermined reference data indicative of at least one defect condition within the vibrating screen unit; and

in response to sufficient similarity with the reference data, identifying the defect condition and estimating an extent to which the defect condition has progressed toward failure.

[0017] Skilled workers in this field will appreciate that this technique provides 'near real-time' feedback on the condition of the screen unit. Any delay in identifying a defect will be associated with the autocorrelation time interval. For example, if the autocorrelation is selected to be a 1 hour interval, the delay between crack initiation and a detectable change in the Z-displacements appearing in the autocorrelation, may be up to an hour. A one hour delay is unlikely to have serious consequences as the majority of defect conditions will take at least 8 hours to progress to failure - sometime days until failure. However, a shorter (or longer) autocorrelation interval may be selected if desired. For example, an autocorrelation period of 4 seconds is possible but this generates far more autocorrelation data than necessary in practical setting. Furthermore, when using short autocorrelation periods (say, less than 60 seconds), it is prudent to wait until several sequential autocorrelations exceed the control limits before initiating any action. Preferably, at least two of the motion sensors are positioned at spaced locations on the vibrating screen unit, and the processor is configured for cross correlation of the output signals from the motion sensors. In a further preferred form, the cross correlation of the sensor outputs is compared to the reference data to determine a likely position of the defect within the screen unit.

[0018] Cross correlation is a measure of the similarity of two output signals as a function of time displacement from each other. Particular cross-correlations between sensors at two or more locations on the vibrating screen unit can indicate a probable location of the crack, and may also provide a better indication of the extent or type of that defect. This gives the screen unit operators the opportunity to control the vibration generator and/or in-feed flow. As discussed above, this can predict the timing of failure to initiate pre-emptive maintenance during scheduled plant shutdowns, and directs the maintenance workers to a likely location of the crack. This avoids the cost and disruption of unplanned shutdowns due to failures and reduces the time required during maintenance.

[0019] In a particularly preferred form, the motion sensors are triaxial accelerometers.

[0020] In some embodiments, the vibrating screen unit has:

a pair of side plates on either side of the at least one screen deck; a feed end configured for connection to an in feed chute;

a discharge end configured for connection to a discharge chute;

a vibration generator for vibrating the at least one screen deck; and a plurality of resilient mounts for supporting the screen unit, such that at least one of the accelerometers is attached to each of the resilient mounts respectively.

[0021] Preferably, the vibrating screen unit has four of the resilient mounts, two of the resilient mounts being attached to each of the side plates respectively such that at least two of the accelerometers are proximate the in-feed end and at least two of the accelerometers are proximate the discharge end, wherein the processor determines cross correlations of the output signals from one or more of: • the two accelerometers proximate the feed end,

• the two accelerometers proximate the discharge end;

• one of the accelerometers proximate the feed end and the accelerometer proximate the discharge end on the opposing side plate; and

· one of the accelerometers proximate the feed end and the accelerometer proximate the discharge end on the same side plate.

[0022] Cross correlation of the 'diagonals' and/or the 'horizontals' may be used to indicate the position of the break (i.e. the defect). For instance equal cross correlations of all diagonals indicates a potential failure near the centre of the screen deck. A cross correlation that is increasing near to the feed end occurs when the front 'horizontal' (i.e. the two accelerometers at the infeed end) cross correlation is greater than the rear the rear horizontal cross correlation.. The diagonal cross correlation may then be used to indicate which side of the front or rear the potential failure will take place. This technique is also used to determine if a side plate has cracked or is loose. The diagonal cross correlation increases significantly as the screen "twisted". This provides a different "finger print" for identifying the beam failure mechanism.

[0023] Preferably, the vibrating screen unit has a single or a series of screen panels extending from the in feed end to the discharge end supported by screen supports attached to support beams extending between the pair of side plates, and the processor is configured to use a cross correlation of a displacement difference between the two accelerometers proximate the in-feed end, and/or a cross correlation of a displacement difference between the two accelerometers proximate the discharge end, to provide an indication of a crack in one or more of the support beams. The use of the cross-correlation values will also provide an indication of the section of the screen in which the crack may be forming (even though it may not provide an indicator of the exact beam with the defect).

Brief Description of the Drawings

[0024] Preferred embodiments of the present invention will now be described by way of example only with reference to the accompanying drawings in which:

Figure 1 is a perspective view of a vibrating screen unit fitted with a defect condition monitoring system according to the present invention;

Figure 1 A shows data collected for a simulated beam with propagating crack recorded at 16Hz and 5g;

Figure 2 is a graph showing the difference in X, Y and Z axis movement of a pair of front motion sensors and a pair of rear motion sensors attached to the spring mounts of a vibrating screen unit with a screen having a cracked support beam;

Figure 3 is a graph showing the Z axis displacement of the motion sensors at all four of the spring mounting of a vibrating screen unit; Figure 4 is a graph of the Z axis displacement of a single motion sensor attached to a front spring mounting of a vibrating screen unit with a screen supported by a cracked beam;

Figure 5 is a graph of a cross correlation between the rear right and rear left Z axis displacement of a vibrating screen unit; Figure 6 is a graph showing the cross correlation of output signals from the front right and front left motion sensors in the Z axis for a vibrating screen unit;

Figure 7 is a graph of the auto correlation of the output signal from the rear left motion sensor of a vibrating screen unit; Figure 8 is a graph of the cross correlation between the Z axis displacement of the rear right and rear left motion sensors in a vibrating screen unit;

Figure 9 is a graph of the differences in X, Y and Z axis displacements between the front and rear pairs of motion sensors on a vibrating screen unit with a loose (defective) side plate

Figure 10 is a screen shot of an orbit analysis, the displacement graph and the frequency graph from the output of motion sensors attached to a vibrating screen unit.

Figure 1 1 is a graph of the displacement data from motion sensors attached to a vibrating screen unit in which a screen has a support beam with a crack propagating to failure;

Figure 12 shows an auto correlation of the output data shown in Figure 1 1 in which an unbiased normalisation technique is applied;

Figure 13 shows the auto correlation of the output data of Figure 1 1 in which a biased normalisation technique is applied; and

Figure 14 is an auto correlation of the output data of Figure 1 1 without a normalisation technique applied. Figure 15 is a basic flow chart showing the operation of the system as it identifies the existence of defects and likely timeline to failure.

Detailed Description of Preferred Embodiments

[0025] Referring to Figure 1 , a defect condition monitoring system 1 is shown applied to a vibrating screen unit 2 used for processing coal. While the system is described here in terms of its use in this application, the skilled worker will understand that this is solely for the purpose of illustration and may be used in other vibrating screens used for processing particulate material.

[0026] Typically a multi-sloped vibrating screen unit (a so called 'banana screen') will have screen displacements from 6mm to 20mm while operating at g-forces in the range of 4g to 5g. These units can weigh up to 32 tonnes with a cyclic sinusoidal motion at a frequency of 12Hz to 18Hz giving rise to high loads on components and support structures. A common failure point is the support beams 6, 7 and 8 attaching the screen deck 5 to the side plates 3 and 4.

[0027] The vibrating screen unit 2 has a pair of side plates 3 and 4 with a screen deck 5 extending transversely between them. The screen deck 5 has supporting beams 6, 7 and 8 rigidly mounting the screening media to each of these side plates 3 and 4 respectively.

[0028] The vibrating screen unit 2 has four spring mountings 10, 1 1 , 12 and 13 rigidly attached to the side plates 3 and 4 respectively such that front spring mounts 10 and 12 are approximate the coal in feed end 15 and rear spring mounts 1 1 and 13 approximate the discharge end 9. A vibration drive 17 vibrates the side plates 3 and 4 which in turn vibrates the screen deck 5 to agitate the coal passing through the unit 2. The vibrating displacement of the unit 2 is accommodated by the springs 14 at each of the spring mounts 10, 1 1 , 12 and 13.

[0029] The monitoring system 1 has motion sensors in the form of triaxial accelerometers 16, 20, 22 and 24 mounted to spring mounts 10, 1 1 , 13 and 12 respectively. The accelerometers produce an output signal of the acceleration (expressed as a g-force) exerted in each of the three orthogonal axes X, Y and Z. Skilled workers in this field will understand that integration of the acceleration provides a velocity and integration of the velocity gives a measure of the displacement in each of the X, Y and Z axes.

[0030] The accelerometers may be rigidly attached in any convenient way however, attachment via high-intensity ceramic magnets allows the sensors to be removed and repositioned relatively easily. Similarly, the accelerometers may be connected to the processor 18 by cable or wirelessly.

Defect Condition Reference Data [0031] Reference data for the vibrating screen unit in various different defect conditions as well as the 'healthy' condition is collected at a range of different vibration frequencies and g- forces.

[0032] Figure 1 A illustrates the collection of reference data for a particular type of defect (specifically, a cracked support beam 6, 7 or 8). The Z-axis displacement from each of the 'rear' (i.e. discharge end) accelerometers 20 and 22 is recorded for a 'healthy' screen, a screen with a loose beam and a screen with a beam that has simulated cracking. Cracking is simulated by progressively cutting through the beam as a percentage of total beam width until failure occurs. Data is recorded at a range of frequencies and g-forces, specifically 14Hz at 4g, 16Hz at 5g and 18Hz at 6g. Figure 1 A shows the data collected for a simulated beam with propagating crack recorded at 16Hz and 5g. The beam crack was tested at 0% cracked (34), 22% (36), 44% (38), 70% (40) and 83% cracked (42). The fracture occurs very shortly after cracked propagation reaches 83%. The reference data reveals that a measureable difference occurs in the difference in Z axis displacement of the rear pair of accelerometers20 and 22, as the crack propagates. This identifies a 'signature' for cracking through the particular beam analysed such that similar sensor outputs from operational plants can be used to flag the presence of the same defect. The propagation rate is tracked to predict the timing of an eventual fracture. In the case of cracked screen support beams 6, 7 and 8 operating under typical conditions, beam fractures are reliably predicted approximately 6 to 8 hours prior to failure. This provides an opportunity to pre-emptively arrange for maintenance such as a scheduled shut down and the necessary technicians and replacement parts.

[0033] Reference data is collected for a range of common defect conditions across all typical operating parameters. This way the ongoing condition monitoring will recognise and flag any defects at an early stage. Crack Propagation in Support Beams

[0034] Referring to Figure 2, the differences between the displacement in the X, Y and Z axes of the front (i.e. infeed end) and rear (i.e. discharge end) pairs of accelerometers 16, 24 and 20, 22 respectively are recorded overtime until beam failure. Dif Fx (44) is the difference in X axis displacement of the front pair of accelerometers (16, 24). Dif Rx (46) is the difference in X axis displacement of the rear pair of accelerometers (20, 22). Dif Fy (48) is the difference in Y axis displacement of the front pair of accelerometers (16, 24). Dif Ry (50) is the difference in Y axis displacement of the rear pair of accelerometers (20, 22). Dif Fz (52) is the difference in Z axis displacement of the front pair of accelerometers (16, 24). Dif Rz (54) is the difference in Z axis displacement of the rear pair of accelerometers (20, 22). The plot shows there is clearly a point at which instability initiates in the screen unit and this increases until eventual failure. The change is most noticeable in the displacement difference in the Z axis between the rear pair of sensors 20 and 22 (i.e. Dif Rz 54).

[0035] Referring to Figure 3, the sensor displacement in the Z axis is recorded for a vibrating screen unit operating at 16 Hz and 5g with a crack in one of the support beams for the screen media. RR Displacement Z (56) is the rear right accelerometer 22. FR Displacement Z (58) is the front right accelerometer 24. RL Displacement Z (60) is the rear left accelerometer 20. FL Displacement Z (62) is the front left accelerometer 16. It can be seen that the Z axis displacement in the rear right accelerometer 22 increases as the crack propagates through the support beam 8 while the Z axis displacement of the rear left accelerometer 20 decreases. These changes become more pronounced until fracture of the beam. Using a cross correlation of the output signals from the rear right Z axis displacement and the rear left Z axis displacement yields a more pronounced effect which is detected earlier which provides greater warning of an impending screen failure.

[0036] Referring to Figure 4, shows the output from a single sensor attached to a screen unit used for dewatering coal (a so-called 'Desliming Screen'). The Z axis displacement is plotted against time until fracture 68 of the screen support beam. The Z axis displacement clearly increases over time and eventually culminates in fracture. The normal operating parameters 64 and 66 are shown either side of the output signal until the Z axis displacement consistently exceeds the upper limit 64 approximately six hours prior (alarm trigger 72) to fracture 68. Prior to the alarm 72, there is some increased instability 70 but not consistently beyond the upper limit 64.

[0037] Skilled workers will note that the two large spikes and drops (74 and 76) in Z axis displacement well before fracture 68, indicate where the vibrating screen unit is shut down and restarted and passes through resonant frequencies (where deflections are large) until all vibration ceases or stabilises. These deflections are beyond the upper and lower thresholds but do not trigger an alarm 72 as they are only momentary.

Z-Axis Auto-Correlation

[0038] An auto-correlation of the output from this single sensor would provide a clear indication of the existence of a defect when the output signals first show faint signs of some instability (whilst still remaining in the upper and lower thresholds 64 and 66). This provides an even earlier warning, well in advance of the alarm 72 when the deflections exceed the upper threshold 64.

Cross Correlation and Auto Correlation of Sensor Outputs [0039] For early failure prediction the processor 18 (see Fig. 1 ) can be configured for autocorrelation and/or cross correlation of the output signals from any of the accelerometers 16, 20, 22 and 24. Auto correlation is carried out on a single sensor only. However, separate auto-correlations may be applied to any one or more of the displacement signals in the X, Y or Z directions

[0040] Figure 5 shows a cross correlation of the Z axis deflections output from the front sensors 16 and 24. However, as discussed above any of the four sensors can be cross correlated with any of the others or with additional sensors applied to the screen unit if desired.

[0041] The algorithm used for the cross correlation analysis may be within the frequency domain, that is instead of a time domain graph showing the signal changes over time (e.g. Figure 4) a frequency domain graph shows how much of the signal lies within each frequency band over a range of frequencies. Conversion to the frequency domain is performed by a transform such as the Fourier transform. In a frequency domain, the signal may be normalised using a biased or unbiased normalisation technique. Alternatively, no normalisation is applied. Figures 5 and 6 are cross correlations of the rear and front sensor pairs respectively in which the signal in the frequency domain has been normalised. In Figure 8, the rear pair of sensors 20, 22 are cross correlated in the frequency domain without normalisation of the signal.

[0042] In Figure 7, an auto correlation of the rear left sensor 20 in the Z axis is shown. Once again, the abnormal operation 80 indicating the presence of a defect is clearly distinguished from the auto correlation of normal operation 78.

[0043] In each case, the cross correlations of the Z axis displacement signals in the frequency domain show data that discriminates between the normal screen operation and the initiation of a defect and therefore potential failure.

Side Plate Defects

[0044] Loose or damaged side plates 3 and 4 are typically characterised in the reference data by excessive movement in the X axis displacement. Referring to Figure 9, the differences in X, Y and Z axis displacements (46 to 52 as per the numbering used in Figure 2) between the front and rear pair of sensors (16, 24 and 20, 22 respectively) are plotted over time. The large differences between the maximum and minimum values for the differences in the X axis displacements (44 and 46) are immediately apparent. The differences between the maximum and minimum values increases as the movement in the side plates increases due to crack propagation or progressive loosening of threaded fasteners holding the side plate. Changes in vibration frequency and unintended orbital motion of the vibrating screen 5 are also indicative of side plate defects.

[0045] Using the present system and auto correlation and/or cross correlation of the sensor output signals, the screen unit operator is made aware of the presence of a defect in the side plates and can take corrective action before the defects seriously deteriorates.

Spring Defects

[0046] Figure 10 shows a screen shot 26 of the graphical user interface at the vibrating screen unit operating system. The three dimensional screen motion or orbit is shown in the window 28 while the vibration frequency spectrum is shown in window 30. The displacement of the motion sensors is plotted in the displacement graph scrolling through window 32.

[0047] Cracks in any of the spring mounts are typically detected in the 3D orbit analysis window 28. However, auto correlation and cross correlation of the motion sensors 16, 20, 22 and 24 may be used to detect a difference between the operation of the springs 14 at each of the resilient mounts 10, 1 1 , 12 and 13. As discussed above, in relation to cross correlation of sensor outputs to detect a cracked support beam 8, the same technique indicates the presence of a cracked spring at an early stage of crack propagation. Otherwise, the spring defect will only come to the operators attention once changes in the screen orbit 28 are observable, or the displacement at the resilient mount where the screen has broken is noticeably reduced (window 32) and extra frequency peaks start to appear in the frequency spectrum 30.

Auto Correlation to Predict Timeline to Failure

[0048] Figure 1 1 plots the motion sensor Z-axis output from a vibrating screen unit in which a screen support beam 8 fails. Data is sampled from the motion sensor area every 30 seconds, then filtered and the displacement plotted. As with Figure 4, the normal operation 78 is punctuated by a transient spike from screen shutdown 74 through a resonant frequency, before resuming operation until eventual failure 68.

[0049] Figure 12 shows the auto correlation of the displacement data using an unbiased normalisation technique, 120 data points are auto correlated such that each correlation is for a one hour period. The number of beam failure indicators is eight and therefore the predicted timeline to failure is 8 hours. In other words, the reference data shows the abnormal operation 80 outside the deflection parameters, begins eight hours before failure 68. The reference data includes auto correlations from 'healthy' (normal operation 78) through to beam failure 68. In the particular beam defect relevant to Figure 12, the reference data shows the first distinct departure from normal operation occurs 8 hours prior to fracture which is then the benchmark timeline for a particular set of operating parameters. "Normal" operation 78 of the screen is indicated by the close correlations evident for the majority of the cross and auto- correlations. As a crack or failure progresses, the correlation moves distinctly away from the norm and results in the increase in the amplitude of the correlation seen in Figures 12, 13 and 14. . Collecting the data for 60 points which is 30 minutes, shows 16 lines outside the norm indicating 8 hour warning and with 30 points which is 15 minutes of data, approximately 30 lines are visible outside the norm which is 7.5 hours warning, but with reduced accuracy due to low number of points used in the calculation. For some screen units, this technique may be used on data that is captured at a greater frequency, for instance every 10 seconds at 1000 samples per second.

[0050] In Figure 13, the auto correlation data is plotted using a biased normalisation of the data, again the distinction between normal operation 78 and the defect indicators 80 is stark with eight clear beam failure indicators plotted prior to beam failure 68.

[0051] In Figure 14, the auto correlation of the motion sensor data has not been normalised and it can be seen that the beam failure indicators 80 are clearly distinguished from the normal operation 78 and once again show eight hours between the first failure indication and eventual failure 68.

[0052] Referring back to Figure 1 1 , the raw displacement data has barely begun to deviate from normal operation 78 at 8 hours prior to beam failure. Hence the use of an auto correlation technique on a single motion sensor provides a readily apparent indicator of a defect at an early stage of the defect propagation. This additional forewarning provides the opportunity to pre-empt the failure 68 through scheduling repairs and replacement parts and/or the necessary maintenance technicians.

[0053] Figure 15 is a flow chart showing the main steps involved in this method of monitoring a vibrating screen unit. Firstly, the accelerometer(s) are attached to the selected location(s), possibly using ceramic magnets for easy re-positioning (100). During operation of the screen, the Z-axis acceleration output is processed (double integrated) to give the Z-axis displacements (1 10). A suitable auto correlation period (say 1 hour) is used to auto correlate the Z-displacements sampled at, say, every 30 seconds (120). Reference data of auto correlated Z-axis displacements is used to determine if the latest auto correlation is departing from normal operation (130). This can be the trigger for an operator alert that operation has become abnormal for a sustained (not transient) period (140). The system processes other axial displacements, optionally auto correlates these displacements and/or cross correlates any of the axial displacements from pre-determined pairs of the accelerometers (150). This is used to cross reference a database of failure 'signatures', defect locations and timelines to failure (160). With this information, the operators have the opportunity to schedule an orderly shutdown process and have an indication of scope of maintenance required. Similarly, the early warning provides a greater ability to adjust the operating parameters (infeed rate, vibration drive etc) to slow the crack propagation and delay failure (170).

[0054] The present invention has been described herein by way of example only. Skilled workers in this field will readily recognise many variations and modifications to these illustrative embodiments, which do not depart from the spirit and scope of the broad inventive concept.