**H G Pisarski, B Hayes, J Olbricht, P Lichter and C S Wiesner**

Paper presented at Charpy Centenary Conference (CCC 2001), Poitiers, France, 2-5 Oct.2001

## Abstract

An idealised Charpy transition curve is provided in Annex J of BS 7910:1999 which enables the 27J transition temperature (T_{27J}) to be estimated from Charpy tests conducted at a single temperature. The validity of this transition curve is examined using data representing a range of ferritic steel grades that includes data from parent plate, weld metal and heat affected zone. It is shown that the idealised Charpy transition curve cannot be relied upon to be accurate or to make consistently conservative predictions of T_{27J}. The main reason for this is that actual Charpy transition curves have a wide range of shapes that cannot be modelled by one curve. Recommendations are made for estimating T_{27J} from tests conducted at a fixed temperature. The implications of using T_{27J} derived from the idealised transition curve to estimate fracture toughness using the Master Curve Charpy-fracture toughness correlation are examined.

## Introduction

Situations arise when there is a need to assess the significance of crack-like flaws in structural components and appropriate fracture toughness data, in the form of fracture mechanics test results, are not available. In these circumstances, Charpy data are often used to estimate fracture toughness through appropriate correlations. Although fracture mechanics, as we know it, was unknown to Monsieur Charpy, we are certain that he would have been pleasurably surprised that the test he developed at the beginning of the last century was still being used to help establish the fracture resistance of structural components.

Annex J of BS 7910:1999 ^{[1]} provides a Charpy-fracture toughness correlation which can be used to estimate fracture toughness in the transition regime of the toughness versus temperature curve for ferritic steels. It is based on the Wallin Master-Curve correlation and requires as input the temperature at which 27J (the 27J transition temperature, T_{27J}) is achieved in standard sized Charpy specimens. However, for components that have already been built or are in operation, Charpy transition curve data are unlikely to be available, so the identification of T_{27J} is a problem. Codes often require that Charpy tests are conducted at a fixed temperature with a target minimum absorbed energy.

In order to provide a means of estimating T_{27J} from Charpy results obtained at a different temperature, Annex J of BS 7910 provides a reference, idealised Charpy transition curve. This curve was derived from a proposal ^{[2]} for defining the limiting thicknesses in steel bridges for the avoidance of brittle fracture. Since the steel specification for bridge construction requires Charpy testing to be conducted at a fixed temperature, an idealised transition curve, referenced to T_{27J}, was used to derive limiting thickness for a range of minimum design temperatures. This curve enables T_{27J} to be estimated from Charpy energies up to 101J. However, the BSI committee preparing Annex J decided that this involved excessive extrapolation and limited the range from 5J to 61J, as shown in *Fig.1*. This paper reviews experimental data to support the idealised Charpy transition curve in Annex J, and suggest an alternative method for estimating T_{27J} where the data does not support it.

## Analysis approach

Charpy transition data for ferritic steels, which included parent plate, weld heat affected zone and weld metal, were collected. To each transition data set, a tanh curve was fitted. This is a mathematical curve that has a characteristic 'S' shape and models the lower and upper shelf as well as the ductile to brittle transition region of the Charpy curve. The parameters of the tanh curve were adjusted to give a fit to each data set with particular attention to the lower shelf to transition regime where T_{27J} lies. In all cases a lower bound fit was made to the data, that is all data points lay on or to the left of the transition curve. Where there were multiple test results at each temperature, a 'best' fit through the middle of data was also made. In addition, these collected data were supplemented by a programme of Charpy testing conducted on one particular ship grade steel plate. This enabled statistical analyses to be conducted on a large, homogeneous data set.

T_{27J} was estimated from the tanh curves fitted to the various data sets and the temperature axis was referenced relative to this temperature. This enabled the fitted transition curves to be compared against the idealised Charpy transition curve shown in *Fig.1*. Other idealisations of the transition curve are also shown in *Fig.1* (from the UK pressure vessel code, BSI PD 5500 ^{[3]} , and from a draft European storage tank code). They are similar to that in Annex J of BS 7910 but have a more limited range. Although these idealisations are not specifically considered in this paper, the comments made are equally applicable.

Further analyses of the data enabled comparisons to be made of T_{27J} from Charpy results obtained at a fixed temperature using the idealised transition curve with T_{27J} estimated from the tanh curve fit to the data. These estimates were made in two ways; using the minimum energy measured at a test temperature, this was typically the minimum of 3 tests, and by using the average energy measured.

## Data collected

Just over 200 Charpy transition data sets were extracted from TWI project reports spanning the past four decades. The data related to specimens notched into parent steel, weld metal, and weld heat affected zone (HAZ), defined as fusion boundary (fb) and general HAZ (excluding fb results). The bulk of the parent steels were C-Mn steels produced to BS1501 (pressure vessel steel) and BS4360 (structural steel). Altogether approximately 2300 Charpy test results were available. *Table 1* below indicates how these results are distributed by steel type and approximate date of manufacture.

**Table 1 Number of Charpy results by steel type and app****roximate date of manufacture**

Year of manufacture or report date | Number of Charpy data points per steel type | Totals per period | ||||
---|---|---|---|---|---|---|

API 5L | ASTM A517 | BS 1501 | BS 4360 | Not defined | ||

1960-1969 | - | - | 78 | 10 | - | 88 |

1970-1979 | - | - | 78 | 305 | - | 383 |

1980-1989 | 12 | - | 106 | 1523 | 50 | 1691 |

1990-1999 | 20 | 8 | - | 80 | 36 | 144 |

Totals per type | 32 | 8 | 262 | 1918 | 86 | Grand total 2306 |

To provide a visualisation for the extent and spread of the data, *Figs 2* and *3* show the transition test results, for parent material and weld metal respectively, shifted by the lower bound 27J transition temperature determined for each data set. In the cases where the data set comprised multiple tests at the same temperature, only the minimum value is plotted.

Equivalent plots for HAZ and fusion boundary results show the scatter to be similar to that of the parent material tests.

In addition to the Charpy data collected from TWI project work, Charpy transition data for a modern Grade A ship steel were available. These data were generated at TWI and consisted of 12 test results measured at each of 13 temperatures, and the results are shown in *Fig.4*.

## Data analysis

When determining T_{27J} from transition test results, the starting point is the construction of a curve. To ensure consistency and reduce variability, the tanh curve was used to describe the transition behaviour. The tanh curve has the form:

[1]

where:

T is the test temperature

T_{o} is the temperature at the midpoint of the transition

A is the Charpy energy corresponding to T_{o}

A+B is the uppershelf energy

A-B is the lowershelf energy

C is a measure of the slope of the transition

It was found that fitting the above expression to the data using a statistical method produced fits that did not model the upper and lower shelves well. Consequently, the parameters were manually varied until a good fit was obtained.

Two sets of tanh curve fits were made to each data set: a 'lower bound' fit, ie with the curve bounding the right of the data, and a 'best' fit where the curve was drawn through the middle of the data. From each of these fits, a value of T_{27J} was determined. The values of T_{27J} were used to reference the data for plotting on a common axis, as in *Figs 2* and *3*.

In addition to determining T_{27J} from the tanh fits, values of T_{27J} were estimated from test results at a fixed temperature using *Fig.1*. These estimates were then compared to the corresponding T_{27J} estimated from the tanh fits for that data.

The ship steel transition data were used to generate simulated data sets which were used in two ways. First, simulated sets of transition test results were generated by selecting one data point at random for each test temperature. A tanh fit was made to each simulated data set and a value of T_{27J} determined. This allowed a distribution of possible T_{27J} values for one material to be built up. Secondly, a simulation was carried out to select random data sets consisting of three results at one temperature. Using these data sets, estimates of T_{27J} using the idealised transition curve were made for comparison with T_{27J} from the tanh curves.

## Results and discussion

The data in *Figs 2* and *3* show, unsurprisingly, significant scatter in the transition and upper shelf regions, particularly for parent steels. Of concern is the scatter in the lower transition region where predictions of T_{27J} are made. In addition, lower shelf Charpy values (typically in the region of 6J) can be and are measured at temperatures more than 30°C below T_{27J}, leading to potentially unsafe predictions of T_{27J}.

*Figure 5* shows the lower bound tanh fits made to data sets (dotted lines) from parent material with the temperature axis referenced against T_{27J} for each data set. It can be seen that the idealised curve (thick black line) gives a reasonable 'average' representation of the bulk of the curves, but not of steels showing sharp transition behaviour.

At temperatures where (T-T_{27J}) > 0; that is, where T_{27J} is estimated from Charpy energies greater than 27J, the idealised curve provides a safe estimate of T_{27J} for shallow transition curves falling to the right of the idealised curve. Conversely, unsafe estimates of T_{27J} are made if the transition curves are steeper and fall to the left of the idealised curve. At temperatures where (T-T_{27J}) < 0, that is where T_{27J} is estimated from Charpy energies less than 27J, the gradient of the transition curve as it approaches the lower shelf is critical, as is the temperature at which the lower shelf is first attained, to whether the predictions from the idealised curve are adequate.

In general, provided that energies not far below 27J are measured, safe estimates of T_{27J} can be made using the idealised curve. Otherwise, potentially unsafe estimates of T_{27J} may be made. Analysis of the transition curves from the parent material, weld metal and HAZ data sets indicates that if the energy is at least 21J but less than 27J, then it is unlikely that the estimated T_{27J} will be non-conservative by more than 5°C. Of course, the error in making a conservative estimate is larger. Unfortunately, because of the steepness of many of the transition curves, it is not possible to make meaningful estimates of T_{27J} from Charpy energies greater than 27J. All that can be claimed with certainty in such circumstances is that T_{27J} will be lower than the temperature at which the Charpy test was conducted.

In practice, the shape of the transition curve is not known when Charpy data are available at only one, or a limited number of temperatures. The errors in predicting T_{27J} from the minimum or average Charpy energies obtained at a single temperature are examined next.

*Figure 6* shows T_{27J} predicted from parent steel Charpy data (from *Fig.2*) using the idealised curve against the T_{27J} value estimated from the lower bound tanh fit to each data set. Only data sets where there were multiple test results at fixed temperatures were used. T_{27J} was predicted from the average value and from the minimum value for each set of test results. *Figure 6* shows that T _{27J} estimated from the idealised transition curve tends to be lower than the T_{27J} found from the lower bound tanh fit, that is the prediction is unsafe. Even if the lowest point at T_{27J} = -10°C is ignored, the error is up to 30°C.

*Figure 7* shows the lower bound tanh curve fits to randomly selected ship plate transition curve data (dotted lines), and compares the results with the idealised transition curve (thick black line). The results are generally similar to the parent steel data, representing various steel grades, shown in *Fig.5*. The mean T_{27J} estimate obtained from this lower bound analysis was -12°C. When the analysis was repeated by employing best tanh curve fits to the same data, the mean T_{27J} estimate was -23°C, indicating a 11°C difference in estimating T_{27J} for this particular steel.

The ship steel data were used to estimate T_{27J} from average and minimum Charpy energies at fixed temperatures (from randomly selected specimens taken in triplicate) using the idealised transition curve given in *Fig.1*. The results are presented in *Fig.8*, where they are compared with the mean ±2 standard deviation estimates from the best tanh curve fits. Interestingly, when estimating T_{27J} from results obtained at temperatures below T_{27J}, the predictions are generally conservative if the mean T_{27J} for the ship plate is being estimated. The likely reason for this is that the ship plate Charpy energies steadily rise from the lower shelf. However, the idealised Charpy transition curve is unsafe if attempts are made to estimate the lower bound ship plate T_{27J} (i.e. highest T_{27J} represented by the mean +2 standard deviation estimate). The situation is improved somewhat if estimates are made using the minimum energy from three Charpy results. However, where T_{27J} is estimated from test results at temperatures above T_{27J}, then non-conservative estimates of T_{27J} are likely, as indicated by *Fig.8* and also *Fig.7*.

_{27J}for a given steel. The reasons for these problems are connected with the fact that Charpy energy measured at one temperature provides no information about the shape of the transition curve, and because actual transition curves have a very wide range of shapes that are not modelled by one idealised Charpy transition curve. These problems are illustrated in

*Figs 9*and

*10*.

Predicting T_{27J} from measured values greater than 27J is safe if the actual (unknown) transition curve is shallower than the idealised curve. It is unsafe if the actual curve is steeper, see *Fig.9*. When predicting T_{27J} from measured Charpy energies of less than 27J, the situation is reversed and the predictions are only safe if the actual Charpy transition curve is steeper than the idealised curve, see *Fig.10*. However, it should be noted that predictions from true lower shelf values (roughly <15J) cannot be made, as such data give no information at all about the transition behaviour.

To sum up, analyses of Charpy transition data from a range of ferritic steel plates, weld metals and HAZs show that, as expected, there is a wide range of transition curve shapes. The idealised transition curve in Annex J of BS 7910 provides an approximately mean fit to the data. Nevertheless, it cannot be shown to provide a conservative lower bound always. Since Charpy tests conducted at a single temperature provide no information as to the shape of the transition curve, potentially unsafe estimates of T_{27J} may be made using the idealised curve. In particular, because of the steepness of many Charpy transition curves, it is not possible to safely estimate T_{27J} from the idealised curve when the Charpy energy is greater than 27J. All that can be safely claimed is that T_{27J} is less than the Charpy test temperature. If the Charpy energy is less than 27J but above 21J, it can be claimed that for each 1J below 27J, (T-T_{27J}) is -1°C, where T is the test temperature. However, even this small correction could be in error by up to 5°C.

## Implication for fracture toughness estimates

As this analysis of Charpy data indicates that an idealised transition curve cannot be relied upon to estimate T_{27J} accurately or conservatively, the question arises as to the implications that this has on estimation of fracture toughness. Although the purpose of this paper is not to examine the Charpy-fracture toughness correlations, some comments are possible. The Master Curve correlation in Annex J of BS 7910:1999 is intended to be conservative. Conservatism is achieved by taking a 95% confidence limit on the basic correlation between the temperature for a median fracture toughness of 100MPa √m (T _{100MPa √m}) for a reference 25mm thick material and T_{27J}. In addition, Annex J recommends that the probability at which fracture toughness is estimated from the Master Curve is the lower 5^{th} percentile (P_{f} = 0.05).

The implications of this can be assessed for the Grade A ship steel plate for which fracture toughness test results are available in addition to Charpy test results shown in *Fig.4*. The results of the fracture mechanics tests are shown in *Fig.11*, where they are compared with the prediction made using the Master Curve correlation as per Annex J. The analyses were made for T_{27J} = -23°C which is the mean estimate from the best tanh fits, for T_{27J} = -44°C, which is the lowest estimate obtained from the idealised Charpy transition curve for tests conducted at single temperatures above T_{27J}, and for T_{27J} = -6°C, which is the highest T_{27J} estimated from the idealised Charpy transition curve, see *Fig.8*.

It is clear that the predicted fracture toughness (K_{J}) transition curve is not very sensitive to the choice of T_{27J}, and that the predictions are very conservative with respect to this particular set of experimental data. Consequently, it can be concluded, at least on the basis of the data available here, that the errors in estimating T_{27J} using the idealised Charpy transition curve are likely to have minimal impact on making non-conservative fracture toughness estimates. The main reason for this is overall conservatism in the fracture toughness correlation. Nevertheless, it is recommended that the best way forward would be to improve/modify the method of estimating T_{27J} in Annex J, possibly along the lines suggested here, and relax the conservatism in the Master Curve by selecting a higher value of P_{f} for making fracture toughness estimates.

## Conclusions

Analyses of Charpy data obtained on ferritic steels which included parent plate, weld metal and HAZ show that the idealised Charpy transition curve in Annex J in BS 7910:1999 is not always conservative. Consequently, it is recommended that it should not be used to estimate T_{27J} from Charpy results obtained at a single temperature. Where energies above 27J are measured at a specific temperature, it is recommended that T_{27J} is assumed to be that temperature. If less than 27J is measured, but more than 21J, it is recommended that T_{27J} is estimated by assuming that for each 1J below 27J, (T-T_{27J}) is -1°C, where T is the test temperature. It is recommended that estimates of T_{27J} are not made when minimum energies below 21J are measured.