WO2003052533A1

WO2003052533A1 - Method and system for on-line monitoring stamping operation

Info

Publication number: WO2003052533A1
Application number: PCT/CN2001/001630
Authority: WO
Inventors: Yangsheng Xu; Ruxu Du; Hang Tong; Ming Ge; Guicai Zhang; Yunyee Leung
Original assignee: The Chinese University Of Hong Kong
Priority date: 2001-12-14
Filing date: 2001-12-14
Publication date: 2003-06-26
Also published as: AU2002218957A1; TW515738B; CN1430721A; HK1053866A1; CN1209597C

Abstract

The present invention is to provide a method for on-line monitoring and diagnosing the stamping process of one or more presses comprises the steps of sampling to a sample under normal stamping condition a plurality of referential strain signals of a frame of a press; establishing a reference curve by smoothing the referential strain signals; setting a tolerance range of the strain of the frame based on the reference curve; sampling an on-line strain signal of the frame by the sensor; determining whether the online signal fall within the tolerance range or not; and controlling the operation of the press in accordance with a result determined in the foregoing step. The invention also provides an on-line system for monitoring and diagnosing the stamping process of one or more press machines includes a device mounted to the press machine for detecting a strain signal thereof; means for sampling a plurality of referential strain signals of the press from the device under normal stamping condition; means for processing the referential strain signals to establish a reference curve; means for setting a tolerance range of the strain of the press based on the reference curve; means for sampling an online strain signal of the press by the sensor; means for determining whether the online signal fall within the tolerance range or not; and means for controlling the operation of the press.

Description

METHOD AND SYSTEM FOR ON-LINE MONITORING STAMPING

OPERATION

FIELD OF THE INVENTION

The present invention generally relates to a method and an on-line system for monitoring and diagnosing the stamping process of one or more presses, particularly to a method and an online system involving wavelet decomposition and reconstruction for monitoring and diagnosing the stamping process of one or more presses.

BACKGROUND OF THE INVENTION

Metal stamping process is a complicated process involving transient elastic and plastic deformation of the sheet metals as well as the static and dynamic behavior of the press. Sensor signals associated with stamping operations are typically transient signals. Hence, the use of FFT spectral analysis method may not work very well. As a result, various transient signal processing methods have been developed, such as short time FFT, and other time domain signal processing methods. For example, Jin and Shi used wavelet transform to compress the tonnage signal into a small set of attributes and then used an Artificial Neural Network (ANN) for classification, Technometrics, November 1999, Vol. 41, No. 4 pp. 327-339. Though, it appeared that the extracted attributes had no explicit physical meaning and hence, monitoring decisions may not be clearly interpreted.

SUMMARY OF THE INVENTION

An object of the invention is to provide a method for monitoring and diagnosing the stamping process of one or more presses, comprising the steps of: a) sampling to a sample under normal stamping condition a plurality of referential strain signals of a frame of a press; b) establishing a reference curve by smoothing the referential strain signals; c) setting a tolerance range of the strain of the frame based on the reference curve; d) sampling an on-line strain signal of the frame by the sensor; e) determining whether the online signal fall within the tolerance range or not; and f) controlling the operation of the press in accordance with a result determined in step e).

Another object of the invention is to provide an on-line system for monitoring and diagnosing the stamping process of one or more press machines, including: a device mounted to a frame of the press machine for detecting a strain signal thereof; means for sampling a plurality of referential strain signals of the press from the sensor under normal stamping condition; means for processing the referential strain signals to establish a reference curve; means for setting a tolerance range of the strain of the press based on the reference curve; means for sampling an online strain signal of the frame by the sensor; means for determining whether the online signal fall within the tolerance range or not; and means for controlling the operation of the press.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a perspective view of a "C" frame cross-shaft press line including an on-line monitoring system for monitoring stamping operation according to the present invention; FIG. 2 is a perspective view of a press configuration in an embodiment mould on blanking or drawing stamping processes shown in FIG. 1 ;

FIG. 3 is a schematic diagram of an instrumentation piezoelectric strain sensor amplifier and an automatic offset correction circuit configured in accordance with an embodiment of the present invention; FIG. 4 is a flow diagram showing operation of the system for monitoring a press during production operation according to the present invention;

FIG. 5 is a partial flow diagram showing operation of the on-line system for monitoring a press in a production cycle according to another embodiment of the invention;

FIG. 6 is a graph plotting the natural frequencies showing the mode shape of a press; FIG. 7 is a graph plotting a waveform showing a strain signal can be decomposed into a number of sections corresponding to different time period in a stamping process;

FIG. 8 is a graph plotting a typical strain sensor signal under three different stamping conditions of the single-stroke stamping operation: (a) normal, (b) workpiece too thick and (c) slug; FIG. 9A through Fig. 9G respectively is a graph plotting a typical strain sensor signal under three different stamping conditions of the continuous stamping operation (a) normal,

(b) workpiece too thick and (c) slug; and

FIG. 10 is a graph plotting a strain signal for single-stroke stamping operation and continuous stamping operation.

DETAILED DESCRIPTION OF THE INVENTION

Without specific indication, the term "tracking threshold" used in the invention has the same meaning as the term "tolerance range". In terms of a physical point of view, a strain in a stamping operation is created owing to the displacement of a press:

_ du _ du _ du du

^{Yx ~ ~}dx ' ^{y = ~}dy^~' ^{xy ~} !bc ^{+ ~}dy^~ where, ^represents the strain, u represents the displacement, and the subscripts x and v describes the directions. It is known that the displacement of the press is related to the force, which in turn is proportional to deformation of a workpiece.

Advanced signal processing needs to extract the features that are related to the conditions of stamping operations. Sensor signals reflecting the strains in the stamping operation are the case. The correlation between the strain signal and the stamping condition may vary depending on various factors such as the type of stamping operations, the speed of the die used (Slot Per Minute), properties of the workpiece and the like. The stamping operation is a transient process. The strain signals are typically transient signals, and can be decomposed into a number of sections corresponding to different time period in a stamping cycle. As shown in Fig. 6, in section 1, the strain is nearly zero since the press is idling at this time. In section 2, the slider of the press moves downwards resulting in a small deformation of the press. As a result, the strain signal increases proportionally. In section 3, the die touches and applies a force to the workpiece. Consequently, the workpiece starts to deform and hence, the strain increases. In section 4, the stamping force surpasses the yield stress of the material and hence, the workpiece breaks and the strain quickly decreases. Finally, in section 5, the die leaves the workpiece and the pressure on the press is released. As a result, the strain varies following the free vibration of the press and gradually damps out. A closer examination indicates that the strain signal first resembles the stamping force (from section 1 to section 4) and then follows the vibration. Hence, it contains both static and dynamic information of the stamping process. As a result, it will be effective for on-line monitoring.

Piezoelectric strain sensor signals have been known to characterize stamping defects during a real time operation of a workpiece. In particular, when the workpiece is too thick, the signal tends to shift left, as shown in sections 2 and 3 in Fig. 6. This is due to the fact that the contact between the dies and the workpiece will occur earlier. If there is a slug, the signal trends to have a larger variation in the end of the stamping. This is due to the fact that the unbalanced stamping force causes a large strain. All these make the piezoelectric strain sensor attractive for on-line monitoring of stamping operations.

However, the features of different failures may occur in different segment of the strain signals. Another difficulty in using strain signals from the machine to monitor the stamping operations is that the collected strain signals are always not smooth enough due to the measurement noise and machine dynamic. Wavelet transform is a newly developed signal processing technique that can decompose strain signal and capture the principal components of the non-stationary stochastic signals encountered in stamping operations. To use the wavelet decomposition may filter out components of a high frequency in the strain signal and only keep components of a low frequency in the same during stamping processes. Based on the principal components of the signal, the "tracking thresholds" can then be constructed by peak value and energy (RMS) value. This is the basic concept of the invention.

Thus, a threshold setting procedure on complex progressive die of continuous stamping processes involving wavelet transform is incorporated in the system and the method of the invention. The wavelet transform decomposes an arbitrary signal x(t) into a superposition of elementary functions obtained by a process of dilating and translating a so-called 'mother wavelet'. The wavelet coefficients obtained from this process can be considered as similarity indices between the section of compared signals and the analyzed wavelet. The continuous wavelet transform is defined as

W (a,b) = ^_r [ x(t)_V/(—)dt

where ψ(t) is a wavelet function, b (b < 5H) is a translation parameter indicating the locality, a (a > 0) is a dilation or scale parameter and '*' denotes a complex conjugation. Continuous wavelet transforms create redundant information. A discrete set of translation and dilation parameters is often sufficient for most tasks. Wavelet transforms within the dyadic framework, i.e. a_} = 2^J and b_{J k} = kl2^j , define an orthogonal wavelet transform.

A function ψ(t) is called an orthogonal wavelet if the family

where m,k e Z, forms an orthonormal basis. The orthogonal wavelet transform then becomes

= r x(t)ψ_m,k(t)dt

The orthogonal wavelet transform represents the most efficient, yet parsimonious representation of the original signal.

Of the strain signals, a piezoelectric strain signal is one of the most commonly used strain signals for a press, particularly for a "C" frame cross-shaft press, in an on-line monitoring of the stamping operation. The process and system according to the invention may be used to control one or more presses simultaneously.

In the system of the invention, one or more strain sensors are mounted to a frame of the press to provide the strain signals. The strain sensors used in the invention may be those conventionally utilized in this art, such as 3D-force sensors, proximity sensors, strain sensors, acceleration sensors, force sensors, and piezoelectric strain sensors. Among these sensors, piezoelectric sensors are preferable.

If the press used is a C frame cross-shaft press, at lease two sensors are mounted to the two columns of the C frame cross-shaft press, respectively. In the method of the invention, a plurality of strain signals are needed to be sampled.

It is obvious that the more samples are sampled, the more accurate the result is. Preferably, the step of sampling is conducted to at least 20 samples. More preferably, the sampling is undertaken to at least 30 samples.

The referential strain signals obtained in step a) of the method of the invention is smoothed by wavelet decomposition and reconstruction to constitute a reference curve.

The steps b) through e) according to one embodiment of the method of the invention may be performed in a computer with Linux C++ program for storage, networking and further processing.

For a single station die, piezoelectric strain signals measured (indirect force measurement) from the press frame are relatively simple (Fig. 10, A). Most operation failures can be detected using the peak-force feature. However, for a multi-station progressive die, strain signals measured from the machine frame are rather complex (Fig. 10, B). As there are multi-peaks in these signals, it is very difficult to monitor the stamping process using only the so-called peak-force (strain). In this case, two-channel strain signals measured from the left frame and the right frame are needed to collect for a more complete description of the process features.

The strain sensors should be placed at such a position in the press that the strain signals can readily be collected. In order to show determine this, a Modal Analysis (MA) experiment has been carried out on a "C" frame cross-shaft press. In the MA experiment has been used a set of 4 sensors with a special software system from National Instrument Inc. From the experiment data, several natural frequencies including 38 Hz, 68 Hz, 140 Hz, and 260 Hz have been found. It is seen that the most sensitive sensor position is at the lower portion (140Hz) of the "C" frame cross-shaft press. This result is expected because it is the weakest spot of the press and is near the stamping operation. The information about the mode shape of the press (The "mode shape" is well-known for those skilled in the art, and is a terminology that describes the natural vibration characteristics of the structure and related to the natural frequency of the structure.) is important for monitoring because it helps to determine the sensor location. The sensor shall be located at where the mode shape reaches its maximum.

In the invention, the strain signal collected from the sensor may further be amplified by an amplifier well known in the art. A charge amplifier is preferable for the invention.

The reference curve of the method of the invention is constituted as follows. The piezoelectric strain sensor is based on physical phenomena that certain crystal structures will emit electrons under a mechanical pressure. One or more significant signal parameters are detected via piezoelectric strain sensors mounted on the "C" frame press in time correlation with the punching processes and is compared with a predefined demand values. A trigger program with an internal trigger signal is made to catch the beginning of the stamping period accurately. Every time the slider of the press moves downwards resulting in a small deformation, the trigger signal is generated when amplitude is around 0.2 N. After the sampling period, unreasonable sample value is omitted by the Grubbs's theory. The Grubbs's Test Method is described as the following:

Suppose the sampling value are X_l5 X₂ , ..., X,. which are arranged in ascending order , then

X- ≤ X₂ -^≤ ... -^≤ X_n, and X; (i = 1 ,2, ... ,n) are in normal distribution.

Then, the mean, X= Based on the statistics theory, Grubbs found out the statistics value distribution is

σ After setting a (normally a =0.05 or 0.01), the Grubbs's value g₀ (n,a) can be found from the table. When p [g_; = g₀ (n,a)]=a is a small probability event, it does not occur when X; obey the normal distribution. Find out the maximum absolute value of υ_t ,

is larger than gO (n,a)

σ ', then the event Xi can be omitted. Repeat the calculations of the mean and variance for

the remaining samples, and compare the remaining maximum absolute υ_i until |t>,| max, is

smaller than gO (n,a) σ '. By using the Grubbs's Test Method, discrepancy value is eliminated one at a time. The calculation is repeated until there is no unreasonable sample value. The computer or manual can adjust the sampling size. The stored value samples, representing the stamping waveform as time and compression force amplitude data, are passed to a processor that statistically fits the data to Grubbs's equation from which the ideal stamping waveform of the stamping event is being monitored. Starting with the internal trigger, the trigger signal will be sampled at a rate of 2-5KHz and the sampling period covers the entire stamping cycle. The monitoring alarm threshold setting is a major concern in monitoring stamping operations. In general, it may be necessary to setup several threshold bands, which adds the complexity for shop floor applications. The sensor signals were captured and sent to the single board computer with Linux C++ program for storage, networking and further processing. The on-line monitoring system according to the invention is an intelligent, reliable, and cost effective system specifically for stamping operations. It increases the efficiency with maximum collision avoidance during superimposed movement of the press and automation system. It will adjust the process automatically during the running of machine at the highest possible precision. It has the following features:

1. It is an intelligent monitoring system. Every time a die is set, it can automatically learn an alarm threshold and diagnose the root cause of the defects. It also can alert users when malfunctions such as slug-pulling, wrinkling, thickness change of raw material, mis-feeding and double-thickness, etc, occur. Since the product defects are minimized, the press and mould will be protected accordingly.

2. It has a stand-alone single board computer. It can be integrated to other press safety devices such as the safety stopper, the die misalignment sensor, and the power monitor. It can also control several stamping machines via network communication system during different process, and record information (e.g., press operating time, number of shots, idling time and etc.) for industry engineers to analysis and improve their product quality and mould design. It is also possible to use one unit to monitoring multiple pressers, or uses a standard PC computer (Min requirement: IBM compatible Pentium) to replace the single board computer. 3. It is a reliable monitoring system. The installation is very easy and it functions equally well for all kinds of C frame cross-shaft press and column frame hydraulic press. The system uses strain sensors mounted onto the frame of the press. It is robust against various kinds of disturbances such as oil, chips, dusts, and the vibration of the nearby press. It has a user-friendly interface with both Chinese and English display on a large LCD screen making the operation easy.

The invention will be further described in combination with the accompanying drawings.

FIG. 1 is a schematic perspective view of a representative "C" frame cross-shaft press assembly 10. In this embodiment, the press line includes a plurality of presses 14-16, each of which can perform a certain metal forming operation on stamping of the material during a production cycle. For example, the presses might sequentially perform the following metal forming operations: continuous stamping operation with progressive die and single-stroke stamping operation. The on-line monitoring system 12 executes a monitoring program for monitoring production operation of individual presses 14-16 based on the wavelet decomposition technology and the data received from the presses, such as from the piezoelectric strain sensors with a charge amplifier. The on-line system also generates process control information, as described in an

LCD display panel and transfers the information to a central control center. In the preferred embodiment of the invention, a monitoring program is designed to allow a user to easily acquire data from an operating press or the central control center. The user interface of the monitoring program makes use of Linux C++ and networking. User selectable commands are organized into menus and submenus that are traversed with the aid of a touch screen or other input device. The monitoring program, preferably written in a Linux C++ computer language, makes use of a variety of libraries for general purposes, data analysis and graphic presentations.

The monitoring program operates on any standard computer, such as IBM compatible machine or Pentium brand programmable computer, equipped with multifunction data acquisition card for signal acquisition. The computer also preferably has at least 128MB of random access memory RAM, a VGA compatible color graphics card, and a LCD display panel. At least 1GB hard disk is preferred as the main storage medium, since much of the monitoring program is very computational oriented. Additionally, information transferred from computer to the central control center includes the location of the press, along with other data regarding the conditions that cause computer to send the signal via communication network. In the case that the central control center is a telnet, all signals and production information are received from the press lines. As shown in FIG.2, a "C" frame cross-shaft press includes an upper binder 20, an upper punch 22, a lower binder 24, and a lower punch 26. The drive motor and flywheel 18, which are generally coupled by a belt drive, a clutch, a brake, and a main drive (for crank presses) with a crankshaft, a connecting rod, and a ram. The deformation process is carried out a fraction (at most one-fourth) of the total cycle for one stroke of the ram. A pull rod is located in the press, extending upward from the drive motor to a crankshaft for moving the upper binder and upper punch in a downward direction. Piezoelectric strain sensor 28 is mounted at where the mode shape reaches its maximum as shown in FIG.6 As shown in FIG. 3, a computer is coupled with a processor 46 and provides an input/output (I/O) interface for the system. A stopper is controlled by output signals from the processor 46 by a press control 40 and a central control center 44. If any one of the stamping defects is detected, the stopper can control the ram movement of the stamping press machine on the top dead point. Thus the processor via the stopper can control the movement of the press and use an alarm light 42 to display the alarm signal on a panel 38. Since the product defects are minimized, the press and mould can be protected accordingly. Two piezoelectric strain sensors 28 are connected to the press via a data acquisition system 34 and a charge amplifier 36 that passes the signals developed by software to an input computer processor 46. Depending upon the type of the dies, various types of 3D-force sensor 30 and fiberoptic laser sensor 32 may be used on monitoring the system. The central control center 44 is adapted to accept signals transferred from processor 46 via communication network and to display the monitoring status of the press machines.

Specifically referring to FIG. 4, a flow diagram shows the operation of the online system of the invention for monitoring and diagnosing a "C" frame cross-shaft press. As shown in the Fig., the system is capable of monitoring and diagnosing a number of features, such as "User information"48, "Sampling parameter" 50, "Alarm set" 52, "Auto learning alarm set"54, "Input samples size"56, "On-line sampling & automatically learn threshold" 58, "On-line monitoring on 24 hour & 7 days" 60, "LCD display" 62, "Use interrupt" 64, "Current statistical report" 66, and "Data transfer" 68. The stamping waveform is repetitively sampled under computer control at a sampling frequency (around 2KHz-5KHz) of the stamping operations. In order to get an accurate stamping waveform, the computer controls the internal trigger level and grain set during different stamping processes. Stored sample size 56 representing the stamping waveform at time and stamping force amplitude data are passed to a processor that statistically fits the data to an equation form 58 which automatically learns an alarm threshold. The on-line diagnosis of FIG. 4 is adapted to diagnose the pressing system for even distribution of the strain signal, depending upon whether the characteristic values Peak Value, (P-V) and energy (RMS) of two piezoelectric strain sensors 28, for all of the press lines are held within the predetermined tolerance ranges determined on the basis of the optimum "tracking threshold". If any one of these two characteristic values is larger than the upper bound (or smaller than the lower bound) of the corresponding tolerance range, the press control system 40 with associated the alarm light 42 will stop the press on the top dead point (TDP). This arrangement facilitates the adjustment of the press, reduces the burden on the operator of the press, and minimizes the reject ratio of the production. The user interrupt 64 is adapted to activate the LCD display 62 for indicating the positions or identification numbers of the abnormal dies. This preferred arrangement permits the operator to easily identify the abnormal die or stamping defects, and facilitates the inspection, adjustment and repair of the die, improving the operating efficiency of the pressing system.

Referring now to FIG. 5, it is a partial flow diagram showing operation of the system for monitoring the press machine according to the invention. In this preferred embodiment of the invention, the stamping waveform is repetitively sampled under computer control at a sampling frequency. The number of samples 80 taken a normal stamping condition is at least 20, and most preferably at least 30. The sampling frequency can be adjusted by the computer in proportion to the press speed, such that the number of samples taken for each stamping event is substantially equal. Stored samples 82 are passed to a processor 86 that statistically fits the data to an equation form 84 to create a "tracking threshold" of upper bound and lower bound 88 by peak value and RMS value. During on-line monitoring process, if any one of these two values for any one of the stamping processes is large than the upper bound or smaller than the lower bound, the press control system 40 associated with the alarm light determines the ram of press on top dead point 92.

FIG. 6 shows the mode shape of the "C" frame cross-shaft press at these natural frequencies. From the figure, it is seen that the most sensitive sensor position will be at the lower portion of the "C" frame cross-shaft press. The strain sensor, 28, shall be located at where the mode shape reaches its maximum.

Referring to FIG. 7, a strain signal is decomposed into a number of sections, corresponding to different time period in a stamping cycle. In SI, the strain is nearly zero since the press is idling. In S2, the slider of the press moves downwards resulting in a small deformation of the press. As a result, the strain signal increases proportionally. In S3, the die touches and applies a force to the workpiece. Consequently, the workpiece starts to deform and hence, the strain increases. In S4, the stamping force surpasses the yield stress of the material and hence, the workpiece breaks and the strain quickly decreases. Finally, in S5, the die leaves the workpiece and the pressure on the press is released. As a result, the strain varies following the free vibration of the press and gradually damps out. A closer examination indicates that the strain signal first resembles the stamping force (from SI to S4) and then follows the vibration.

Now referring to FIG. 8, it shows strain sensor signals under three different stamping conditions of the single-stroke stamping operation: (a) normal 104, (b) workpiece too thick 106, and (c) slug 108. Based on the figure the following observations can be made: First, the signal is sensitive to the stamping conditions. In particular, when the workpiece is too thick, the signal tends to shift left (in S2 and S3 of FIG.7). This is due to the fact that the contact between the dies and the workpiece will occur earlier. Second, if there is a slug, the signal trends to have a larger variation in the end of the stamping.

The monitoring system of the present invention offers several important advantages distinguishing it form conventional monitoring apparatus. As shown in FIG. 9, the threshold curve set using wavelet decomposition technology is a kind of "tracking threshold". The threshold setting equation form 58 procedures are as follows:

1. Get 20-30 samples under normal stamping condition, for every sample we get two strain signals from left column and right column of the press, respectively. 2. Using wavelet decomposition and reconstruction, we get a filtered low-frequency component of every strain signal. This filtered strain signal is smooth and can be used to set overall threshold.

3. Calculate the mean curve (m,(t) for left strain signals and m_r(t) for right strain signals) and standard deviation curve (v,(t) for left strain signals and v_r (t) for right strain signals) of the 20-30 filtered strain signals. Plot m_l (t) +3 v_z (t) and m_t (t) -3 v, (t) , we get upper threshold and lower threshold for

left strain signals 110; Plot m_r(t) +3 v_r (t) and m_r (t) -3 v_r (t) , get upper "tracking threshold" bound and lower "tracking threshold" bound for right strain signals, 112. See Fig. 9-A.

Fig. 9-B shows using wavelet decomposition to create a "tracking threshold" under normal condition 114, 116.

Referring to Fig. 9-C through Fig. 9-G, using wavelet decomposition to create this curve threshold, the production operation failures of the continuous stamping operation such as material variations, misfeed 118, 120, slug pulling 122, 124, 126, 128, 130, 132, 134, 136, double hit, material end, etc can be detected. Generally, different process failures have different features, and these features can be found in different segments of the strain signals. However, to divide the strain signal into multi segments according to a certain process is very difficult. Using this approach, the monitoring system can easily find the process change occurred in different segments during the continuous stamping operation with high precision progressive die.

For the single station die, the strain signals measured (indirect force measurement) from the press frame are relatively simple (Fig. 10- A) 138. Most operation failures can be detected using the peak-force feature. However, for a multi-station progressive die, the strain signals measured from the machine frame are rather complex (Fig. 10- B) 140, 142. In this case, two-channel strain signals are necessary for a more complete description of the process features. The features of different failures may occur in different segments of the strain signals (indirect force). Although the present invention has been described and exemplified in terms of certain preferred embodiments, other embodiments will be apparent to those skilled in the art. The invention is, therefore, not limited to the particular embodiments described and exemplified, but is capable of modification or variation without departing from the spirit of the invention, the full scope of which is delineated by the appended claims.

Claims

What we claim is:

1. A method for monitoring and diagnosing the stamping process of one or more presses, comprising the steps of: a) sampling to a sample under normal stamping condition a plurality of referential strain signals of a frame of a press; b) establishing a reference curve by smoothing the referential strain signals; c) setting a tolerance range of the strain of the frame based on the reference curve; d) sampling an on-line strain signal of the frame by the sensor; e) determining whether the online signal fall within the tolerance range or not; and f) controlling the operation of the press in accordance with a result determined in step e).

2. The method of claim 1, wherein said referential strain signals come from a strain sensor mounted to a frame of the press.

3. The method of claim 2, wherein said strain sensor is a piezoelectric sensor.

4. The method of claim 1 or 2 or 3, wherein said sampling a plurality of referential strain signals is conducted to at least 20 samples.

5. The method of claim 1, wherein said smoothing said referential strain signal in step a) is implemented by wavelet decomposition and reconstruction.

6. An on-line system for monitoring and diagnosing the stamping process of one or more press machines, including: a device mounted to the press machine for detecting a strain signal thereof; means for sampling a plurality of referential strain signals of the press from said device under normal stamping condition; means for processing said referential strain signals to establish a reference curve; means for setting a tolerance range of the strain of the press based on said reference curve; means for sampling an online strain signal of the press by said sensor; means for determining whether said online signal fall within said tolerance range or not; and means for controlling the operation of the press.

7. The on-line system of claim 6, wherein said device comprises one or more strain sensors.

8. The on-line system of claim 7, wherein said device comprises a signal amplifier.

9. The online system of claim 7, wherein said strain sensor is a piezoelectric strain sensor.

10. The on-line system of claim 8, wherein said signal amplifier is a charge amplifier.

11. The on-line system of claim 6, wherein said processing referential strain signals to establish a reference curve is implemented by wavelet decomposition and reconstruction.

12. The on-line system of claim 6, wherein said press is a C frame cross-shaft press comprising two columns.

13. The on-line system of claim 12, wherein said frame comprises at least two strain sensors mounted to the two columns, respectively.

14. The on-line system of claim 13, wherein said strain sensors are piezoelectric strain sensors.

15. The on-line system of claim 7, wherein said strain sensors are located where the mode shape of the press reaches its maximum.